Fix nested fenced code block parsing

[wesm]

Replaces the lazy fenced-code regex in transcript parsing with a scanner that tracks the opening backtick fence length, so shorter nested fences remain inside the outer code block.

Adds a regression case for a markdown fence containing a nested qmd fence.

Fixes #721.

[roborev-ci]

roborev: Combined Review (7d374e3)

Medium issue found in the fenced code parsing change.

Medium

• frontend/src/lib/utils/content-parser.ts:210 - closingFence treats any backtick run at least as long as the opener as the closing fence, even when it appears inline or has non-whitespace text after it. This can prematurely end a fenced code segment when the code contains a string/comment with ``` or ```` before the real closing fence, causing the rest of the code to be parsed as normal text or tool/thinking blocks.
  • Fix: Only accept closing fences at a valid fence line start with up to three leading spaces and only spaces/tabs after the backticks. Add regression coverage for same-length backtick runs inside code content.

---

Panel: ci_default_security | Synthesis: codex, 10s | Members: codex_default (codex/default, done, 2m23s), codex_security (codex/security, done, 1m0s) | Total: 3m33s

[roborev-ci]

roborev: Combined Review (80f01ae)

No issues found.

---

Panel: ci_default_security | Synthesis: codex | Members: codex_default (codex/default, done, 4m50s), codex_security (codex/security, done, 1m25s) | Total: 6m15s