Fix #491: cache C# syntax trees to eliminate repeated full-document parses

[clrudolphi]

🤔 What's changed?

• Added ICSharpSyntaxTreeCache, a shared, thread-safe, MRU-capped cache for parsed C# syntax trees, and wired it into ScenarioTestTargetResolver and CSharpAttributeLiteralResolver. Both previously called CSharpSyntaxTree.ParseText(File.ReadAllText(...)) on every single invocation, re-parsing the same generated .feature.cs/step-definition files from scratch on every reqnroll/resolveTestTargets or rename request.
• Added RunTestCodeLensResultCache to de-duplicate concurrent calls to RunTestCodeLensService.GetTargetsAsync for the same file. That method resolves every scenario in a .feature file, but was being invoked independently by the tagger and by every visible Scenario line's own out-of-process CodeLens data point — on a large file that meant N+1 concurrent full-document walks, slow enough that individual callers hit VS's own CodeLens timeout and the lens got stuck on "Loading data...". Cancellation is deliberately decoupled per caller (via AsyncLazy<T>) so one caller giving up never aborts the shared computation for the others.
• Changed the Run CodeLens title to "▶ Run Scenario" / "▶ Run Scenarios" (singular vs. plural for Scenario Outlines), which required threading a new Detail field through GherkinDocumentSymbolService/GherkinNavigationBarSymbolService (Scenario vs. Scenario Outline can't be told apart from Kind alone) and a new IsScenarioOutline flag through RunTestTargetEntry.
• Added logging across the previously-uninstrumented OOP-to-in-process CodeLens callback boundary (RunTestCodeLensCallbackListener, RunTestCodeLensDataPointProvider, RunTestCodeLensDataPoint) — this is what surfaced the VS-timeout root cause behind the result-cache fix above.
• Fixed LspInterceptingPipe so owned-RPC request/response traffic (used for the extension's own shutdown handshake, etc.) is run through the send/receive interceptors, closing a blind spot in LspInspectorLogger where that traffic was previously invisible.
• Updated docs/LSP-IDE-Support-Architecture.md and docs/Test-Runner-Integration-Design.md to reflect the new cache and the corrected cost/fix note.

⚡️ What's your motivation?

Fixes #491. On the VeryLargeFeature stress-corpus solution (~1,300+ scenarios), a single .feature tab pegged a CPU core and Run Scenario CodeLens never resolved. Root-caused via live VS Experimental-instance testing plus runtime log analysis:

1. Every reqnroll/resolveTestTargets call re-parsed the whole generated .feature.cs file from disk with no cache — confirmed via before/after log measurements taking it from ~153ms to ~3.6ms per call.
2. Fixing (1) exposed a second bottleneck: with N visible Scenario CodeLens lines each independently triggering their own full-document walk, the combined load still took long enough (~26-30s) that VS's own per-data-point CodeLens timeout fired before the walk finished, throwing OperationCanceledException and leaving the lens stuck.

Verified fixed via two full live VS Experimental-instance test runs against the same corpus: inlay hints in a few seconds, Run Scenario CodeLens resolving in ~30s (a single shared walk instead of N concurrent ones), label correctly pluralized, and the Run popup/test execution working end to end. No OperationCanceledException in either run's logs.

🏷️ What kind of change is this?

• 🐛 Bug fix (non-breaking change which fixes a defect)

🧩 Area(s) touched

• LSP server (src/LSP)
• Visual Studio extension (src/VisualStudio)

♻️ Anything particular you want feedback on?

RunTestCodeLensResultCache's cancellation model: each caller's own token only governs how long that caller waits (backed by AsyncLazy<T>'s documented contract), never the underlying shared computation — a 60s independent timeout bounds the computation itself instead. Flagging this in case there's a simpler pattern preferred for this kind of shared/de-duplicated async work elsewhere in the codebase.

📋 Checklist:

• I've changed the behaviour of the code
  • I have added/updated tests to cover my changes.
• My change alters as-built behaviour described in a design doc under docs/
  • I have updated the relevant doc accordingly (see docs/AsBuilt-Reconciliation-Reminder.md).
• Users should know about my change

🤖 Generated with Claude Code