export: add C ABI wrappers for paragraph measure/split + language-aware font parse

[carlos7ags]

Summary

Surfaces four v0.8.0 Go-side additions to the C ABI so non-Go callers (Python, Node, Rust, etc.) get the same surface area without each having to write their own CGO bridge.

New C export	Wraps	Source PR
folio_paragraph_measure_lines(p, max_width)	layout.Paragraph.MeasureLines	#238
folio_paragraph_measure_height(p, max_width)	layout.Paragraph.MeasureHeight	#238
folio_paragraph_split_after_line(p, n, max_width, *head, *tail)	layout.Paragraph.SplitAfterLine	#253
folio_font_parse_for_language(data, length, lang)	font.ParseFontForLanguage	#277

Out of scope

font.Fallback / layout.NewParagraphFallback (#225) is intentionally NOT wrapped here. Per its PR description, C ABI exposure is Phase 2 work tied to promoting *EmbeddedFont to an interface so Fallback becomes a drop-in replacement at every existing call site (heading, link, list, tab, table cell, page-level text, C ABI). Wiring the C ABI to the Phase-1 *EmbeddedFont-only constructor and then refactoring it again in Phase 2 would create churn for downstream consumers.

html.Options.Logger (*slog.Logger), Options.Client (*http.Client), and Options.StrictAssets are Go-specific by nature — slog and *http.Client don't have portable C equivalents. Adding a C-flavoured logging callback / HTTP transport surface is its own design pass, deferred.

API design notes

• Split semantics for split_after_line: returns two handles via out-pointers. Either may be 0 for the no-op halves: n <= 0 returns (0, full clone); n >= total lines returns (full clone, 0). The receiver is unchanged. Caller frees non-zero handles via folio_paragraph_free. out_head and out_tail themselves must be non-NULL — the wrapper returns errInvalidHandle if either is missing rather than crashing.
• measure_lines returns int32_t: -1 on bad handle. Callers should check folio_last_error for the message.
• measure_height returns double: -1.0 on bad handle (negative height is impossible for a real paragraph).
• font_parse_for_language accepts NULL lang: treated as empty string, which font.ParseFontForLanguage handles by falling back to face 0 (matching folio_font_parse_ttf semantics for non-TTC inputs).

Tests

test_cabi.c grows from 397 to 411 assertions (+14 new), covering:

• measure_lines happy path (narrow vs wide width invariant), bad handle
• measure_height happy path, bad handle
• split_after_line n=2 success path with head/tail line-count assertions
• split_after_line edge cases: n=0 (head=0, tail=full), n=large (head=full, tail=0)
• split_after_line bad handle, NULL out_head, NULL out_tail
• font_parse_for_language NULL data rejection (round-trip with real TTC is covered by Go-side tests in font/load_test.go)

Audit

scripts/audit-cabi.sh --build reports:

• Go //export directives: 393
• Header declarations: 393
• Built dylib symbols: 393

All three in sync.

Test plan

• go vet ./export/... — clean
• make test-cabi — 411 passed, 0 failed
• scripts/audit-cabi.sh --build — clean