Lift TextBox caret glyph-advance fast path into IFormsText

[vchelaru]

Fixes #3542.

Problem

TextBoxBase.GetIndex (caret hit-testing) had an #if XNALIKE fast path (BitmapFont XAdvance walk, allocation-free O(n)) and an #else fallback (repeated growing-substring MeasureString, O(n^2), allocating a substring per character). Since Forms controls moved into GumCommon (which defines no XNALIKE), every backend except FRB's XNA-family builds silently lost the fast path and took the O(n^2) fallback — including standalone MonoGame/KNI/FNA.

Change

• Added IFormsText.GetCharacterAdvance(char), a default interface method that falls back to measuring a single-character string (O(1) per call → O(n) total, already an improvement over the old growing-substring fallback).
• The XNA-family concrete Text (RenderingLibrary/Graphics/Text.cs) overrides it with the allocation-free BitmapFont.XAdvance lookup — the same computation the old #if XNALIKE block did.
• TextBoxBase.GetIndex now calls coreTextObject.GetCharacterAdvance(...) directly with no #if — virtual dispatch on the concrete Text instance picks the right implementation at runtime. Result: MonoGame/KNI/FNA/FRB get the allocation-free walk back (restoring what was silently lost), and Raylib/Sokol move from O(n^2) to O(n).

Verification

• MonoGameGum.Tests: full suite green (1821/1821), including the existing FontScale_ShouldScaleClickHitTestX regression test (TextBox odd/broken behavior #2687) which now exercises the new GetCharacterAdvance override.
• RaylibGum.Tests: full suite green (455/455), including a new GetCaretIndexAtPosition_MapsFurtherRightClicksToLaterIndices test pinning the new default-interface fallback path (previously untested — Raylib's caret hit-test had no coverage at all). Needed adding RaylibGum.Tests to GumCommon's InternalsVisibleTo list (mirroring the existing MonoGameGum.Tests entry) to reach the internal GetCaretIndexAtPosition.
• SkiaGum build: clean (unaffected — Skia doesn't compile TextBoxBase.cs yet).
• FRB canary (FlatRedBall.Forms.DesktopGlNet6.csproj), built from the primary checkout: baseline (main) clean, branch clean, 0 new errors/warnings.

Manual test: built Samples/GumFormsSample/MonoGameGumFormsSample (clean, no sample edits needed — FrameworkElementExampleScreen already has TextBox instances) and opened the .sln for a real click-to-caret sanity check, given this touches a genuinely user-facing input path (clicking in a TextBox) across runtimes, even though the unit tests directly cover the click→index computation end-to-end.

Steps to verify manually in the opened solution:

1. Run the MonoGameGumFormsSample project (F5).
2. Press 2 to switch to the Framework Element Example screen (only works while no control is focused — click empty space first if needed).
3. Click into either TextBox at various horizontal positions and confirm the caret lands at the clicked character, including after typing enough text to test near start/middle/end.

🤖 Generated with Claude Code

---

Additional fix (found during manual testing above)

Pressing 2 in the MonoGameGumFormsSample to reach FrameworkElementExampleScreen crashed with:

System.NullReferenceException: Object reference not set to an instance of an object.
   at Gum.Forms.DefaultVisuals.DefaultButtonRuntime..ctor(Boolean fullInstantiation, Boolean tryCreateFormsObject)

Root cause: the sample directly constructs a legacy V1 default-visual class (new DefaultButtonRuntime() — the documented "create a control through its visual type" pattern). DefaultButtonRuntime's constructor reads Styling.ActiveStyle.Colors.Primary, but Styling.ActiveStyle (the V1/V2-shared style) is only ever set inside FormsUtilities.InitializeDefaults's V1/V2 switch cases. The sample only calls InitializeDefaults with DefaultVisualsVersion.Newest (V3), which populates a separate DefaultVisuals.V3.Styling.ActiveStyle and never touches the legacy one — so it stayed null for any V3-only app that also directly instantiates a legacy visual class.

Fix: Styling.ActiveStyle now lazily self-initializes (loading the embedded UISpriteSheet) when read while unset, instead of depending entirely on InitializeDefaults(V1/V2) having run. Extended the constructor's null-spritesheet fallback (already present for XNALIKE) to RAYLIB and SOKOL too, so the lazy path works on every platform.

Verification:

• New regression test ButtonTests.DefaultButtonRuntimeConstructor_ShouldNotThrow_WhenActiveStyleNeverInitialized — confirmed it reproduced the crash before the fix, green after.
• MonoGameGum.Tests (1822/1822), MonoGameGum.Tests.V2 (120/120), MonoGameGum.Tests.V3 (59/59), RaylibGum.Tests (455/455) all green.
• SkiaGum build clean. SokolGum build fails on pre-existing, unrelated native-binding errors (sg_image/sgp_blend_mode not found — missing submodule, not touched by this change).
• FRB canary (FlatRedBall.Forms.DesktopGlNet6.csproj): baseline clean, branch clean, 0 new errors.
• Sample rebuilt clean with the fix applied — ready for you to re-verify by pressing 2 again in the already-open Visual Studio solution.

---

Sample cleanup (follow-up to the crash fix above)

The crash above only surfaced because the sample demonstrated the "create/subclass a control through its visual type" pattern using obsolete V1 classes (DefaultButtonRuntime, DefaultScrollViewerRuntime) while the rest of the sample runs V3. Migrated the sample off V1:

• FrameworkElementExampleScreen: DefaultButtonRuntime → V3.ButtonVisual; the custom ScrollViewer background template moved from DefaultScrollViewerRuntime/ColoredRectangleRuntime → V3.ScrollViewerVisual/NineSliceRuntime (V3's Background child type).
• StyledButtonRuntime (dead sample code — never actually instantiated, but still triggered a CS0618 warning via its base type): now subclasses V3.ButtonVisual and sets BackgroundColor directly, since V3 computes highlighted/pushed/etc. tints from that one color instead of requiring per-state variable edits.

Verified: MonoGameGumFormsSample.csproj and the full MonoGameGumFormsSample.sln build clean (0 errors), and the CS0618 obsolete-API warnings from these two files are gone on a full rebuild.

---

Second crash fix (still on screen 2)

Pressing 2 still crashed after the first Styling.ActiveStyle fix, this time in SliderVisual.SetValuesForState:

System.NullReferenceException: Object reference not set to an instance of an object.
   at Gum.Forms.DefaultVisuals.V3.SliderVisual.SetValuesForState(Boolean isFocused, Boolean isEnabled)

Two distinct bugs, both surfaced by the same root cause: FrameworkElementExampleScreen's constructor overrides FrameworkElement.DefaultFormsTemplates[typeof(Button)] with a custom visual for its own demo button — and that override leaks into every composite control built afterward that internally constructs a Button sub-part, like Slider's thumb.

1. Gum.Forms.DefaultVisuals.V3.Styling.ActiveStyle had the exact same bug as the legacy Styling fixed earlier in this PR — it's a separate static class, so that first fix didn't cover it. Applied the identical lazy-self-init fix.
2. SliderVisual.ThumbInstance is nullable by design (the as ButtonVisual cast fails when the registered Button template isn't a ButtonVisual — exactly what the screen's override does), but SetValuesForState dereferenced it unguarded. Added the null check. Checked every other V3 visual that builds a nested Button (ScrollBarVisual) — it never dereferences its equivalent nullable fields unsafely, so this was the only latent instance of the bug.

Verification:

• New regression test SliderTests.UpdateState_ShouldNotThrow_WhenButtonTemplateIsNotButtonVisual reproduces both bugs (registers a non-ButtonVisual Button template, constructs a V3 Slider) — confirmed it failed on each bug in turn before its respective fix, green after both.
• MonoGameGum.Tests (1823/1823), .V2 (120/120), .V3 (59/59), RaylibGum.Tests (455/455) all green. SkiaGum build clean.
• FRB canary: baseline clean, branch clean, 0 new errors.
• Sample rebuilds clean — ready for another pass at pressing 2.