🐛 FIX: myst_role backslash-guard misparse at inline start

[chrisjsewell]

What

Fixes #62: myst_role misparsed inline content that starts with a role and ends with a backslash:

md.render("{foo}`ar`\\")
# before: <p>{foo}<code>ar</code>\</p>          (role not captured)
# after:  <p><code class="myst role">{foo}[ar]</code>\</p>

Why

The "starting backslash escape" guard read state.src[state.pos - 1]. At state.pos == 0 that is Python negative indexing — it reads the last character of the inline source (the except IndexError never fires), so a role at the start of the inline was rejected whenever the content ended with \.

The guard is removed entirely rather than bounds-checked, because it was wrong in both remaining respects too:

• For its intended case (\{foo}x``) it was dead code: { is CommonMark-escapable, so the core `escape` rule (which runs before this rule) consumes `{` wholesale — `myst_role` is never invoked there. Verified unchanged before/after.
• For an escaped backslash (\\{foo}x``) it wrongly suppressed a live role; it now renders a literal \ followed by the parsed role, matching the CommonMark analogy (`\a` → `<em>a`). This is the "edge case `\{`" the guard's own comment flagged as broken.

This mirrors the analysis done for the section_ref plugin in #144, where the same guard pattern was dropped.

Testing

• 3 new fixture cases: the myst_role misparses trailing \ #62 repro, \{foo}x (still escaped, via the core rule), and `\\{foo}`x (now a live role).
• Adversarially probed old-vs-new behavior across commonmark/default/gfm presets, table cells, link labels, image alts, and emphasis adjacency: the only differences are the two intended fixes. Full suite 511 passed; ruff and strict mypy clean.
• Sole caveat: in a zero-preset config without the core escape rule, \{foo}x`` now parses the role (previously suppressed-but-still-literal-backslash). No realistic consumer runs myst_role without the escape rule; a code comment now documents the assumption.

Note: the same latent state.src[state.pos - 1] wraparound exists in dollarmath (only reachable with the non-default allow_digits=False) — left for a separate PR.

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 92.90%. Comparing base (d11bdaf) to head (581fefe).
⚠️ Report is 15 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #145      +/-   ##
==========================================
+ Coverage   92.80%   92.90%   +0.10%     
==========================================
  Files          31       40       +9     
  Lines        1835     2284     +449     
==========================================
+ Hits         1703     2122     +419     
- Misses        132      162      +30     

Flag	Coverage Δ
pytests	92.90% <ø> (+0.10%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.