fix(pe3): treat ignore-list code fences in Markdown as ignore-file entries

[jwcastillo]

Problem

#787 stopped PE3 from flagging single-pattern lines of real ignore files. The same lines shown in a Markdown code block labelled as an ignore list still raise HIGH PE3 (issue #830):

```gitignore
# Environment files
.env
.env.local
```

Fix

static_patterns_privilege_escalation.py now collects the body ranges of fenced blocks whose info string starts with an ignore-list language: gitignore, git-ignore, ignore (GitHub Linguist's "Ignore List" aliases) or dockerignore, case-insensitive, first word only. It does this for Markdown and text files only. A match inside such a block goes through the existing _is_ignore_file_exclusion_entry check, exactly like a line of a real ignore file.

Still flagged:

• comments, ! negations, commands, host paths and any line that is not one pattern token;
• unfenced lines and lines after the fence;
• other fence languages (bash, text, yaml, no language, gitignore-like, ...);
• the same fence inside a non-Markdown file, such as a Python docstring.

Tests

tests/nodes/analyzers/test_pe3_ignore_file_entries.py gets a new TestIgnoreListCodeFencesInMarkdown class with 33 tests:

	on main @ 4c52fbb	this branch
exempt cases (ignore-list fences in SKILL.md / README.md / references, tilde fence, unclosed fence)	21 fail	pass
fail-closed controls (other languages, negation, comment, command, host paths, prose after fence, script file)	12 pass	pass

• make lint and make format pass.
• pytest -m "not integration and not provider" tests/ gives 13702 passed, 20 skipped, 4 xfailed and 2 failed. The 2 failures are test_suppression.py::test_dump_baseline_concurrent_writers_publish_complete_documents[.yaml/.json], which failed under load in the full serial run. Rerun on their own, they pass on both this branch and main.

Real-world check

skillspector scan skills/git-workflow --no-llm on affaan-m/ECC:

• main reports PE3 on lines 522 and 523 (.env and .env.local inside the ```gitignore block) and on line 669.
• This branch reports only line 669, which is git add .env # Contains API keys in a "BAD" example. It is a command, so it should stay flagged.

Fixes #830. Refs #644.

Prepared with Claude Code; I reviewed and tested the change.

[yashrajp22]

One remaining integration issue: the ignore-list exemption is lost across production scan windows. Confirmed in source and a freshly built wheel; the inline comment includes the boundary case.

[yashrajp22]

Could we preserve the ignore-list language across scan windows? The production runner rebuilds an open fence using only its marker and width, so later windows lose the gitignore label this helper needs. I reproduced this with a 256,001-character Markdown file containing one gitignore block and .env near the end: direct analyze() returns no PE3, but the runner and CLI still report it at line 36,571, in both source and a fresh wheel. The 256,000-character control passes. Please carry the language into the window context and add a regression check through the production runner.

[jwcastillo]

Thanks @yashrajp22, confirmed and fixed in cdc7599.

Cause: _markdown_context_prefix rebuilt an open fence as marker * width + "\n", so every window after the first saw a bare fence and the gitignore label was gone.

Fix (static_runner.py): the fence state and the open/close transitions now also record the opening info string, and the synthetic opening line includes it (```gitignore). This applies to all three prefix cases: inside a fence, a window starting on the opening line, and a window starting on the closing line. Findings inside the synthetic prefix still map before the owned range, so the prefix itself does not add findings.

Regression checks through run_static_patterns (TestIgnoreListFenceAcrossScanWindows):

• a ```gitignore fence longer than SECURITY_VIEW_WINDOW_CHARS with `.env` in a later window: no PE3 (failed before the fix at line 36574);
• the same file with a ```bash fence: PE3 still reported on the .env line.

CLI check on a 256,001-character SKILL.md with one gitignore block and .env at the end: no PE3 on this branch.