fix(docs): preserve heading and anchor fidelity

[matcra587]

Description

Heading lookup could miss published anchors or mistake code inside fences for a section. Share source-range heading parsing between extraction and section search, preserve literal source bytes and resolve exact anchors, including duplicate suffixes, inline markup, autolinks and Unicode text. List-contained fences retain their boundaries.

Review Focus

• headings.go and extract.go: heading ranges, visible text and exact anchor selection.
• fence.go: fence characters, lengths and list indentation; focused tests cover false headings and container termination.
• cursor.go: reject older cursor positions after section-selection semantics change.

Test Procedure

• mise run ci: formatting, lint, offline race tests, vulnerability checks, release checks and all five compilation targets.
• Named regression assertions first failed for duplicate, punctuation, inline-link and autolink anchors, fence lengths and list-contained fences; all pass after the fixes.
• Race-enabled live stdio MCP canary across all nine languages; 30-second FuzzExtractHeading run.

Pre-flight Checklist

• Changes are limited to a single feature, bugfix or chore
• Tests pass and code is formatted/linted
• Docs updated (if needed)

Additional Notes

Cursor format is now v3. Clients holding earlier cursors must restart their original selection. Existing initial calls and byte offsets remain supported. Parsing remains bounded rather than implementing a general Markdown renderer.