fix: dice regex matches bare D1/D0 in sourcebook text (#1126)

[0xguy07]

Summary

Restrict the bare-die dN/DN form in DICE_REGEXP to N=2–9 or N with two-or-more digits, so impossible bare dice like D0/D1 no longer hijack sourcebook map labels (e.g. A2, B1, C1, D1 in Vecna: Eve of Ruin). Explicit-count forms (1d1, 2d0, …) are unchanged. A standalone npm test script is added covering the bug and existing-behavior regressions.

Type of change

• 🐛 Bug fix (non-breaking change that fixes an issue)
• ✨ New feature (non-breaking change that adds functionality)
• ♻️ Refactor (no functional change)
• ⚡ Performance improvement
• 🧪 Tests (adds or updates tests)
• 📝 Documentation
• 🔧 Build/CI
• 🚨 Breaking change

⚠️ AI usage disclosure (required)

• I did not use AI for this PR.
• I did use AI for this PR (describe below).

What was AI-assisted: The regex change in src/common/utils.js and the test cases in test/dice_regex.test.js were drafted with Claude (Anthropic, Opus 4.7). I (the PR author) reviewed every change, ran the bundled test script, and confirmed the behavior on the issue's reproducer sentence and on standard dice expressions. No build/CI tooling, documentation, or unrelated code was modified by AI.

This is a resubmission of #1375 — the prior PR was closed after @dmportella asked about AI use; this one discloses it upfront and adds the test coverage that the prior thread noted regex changes need.

Motivation & context

• Related issue(s): (Minor) Replacement of D1 false positive #1126
• Prior PR: fix: dice regex matches bare D1/D0 in sourcebook text (#1126) #1375 (closed without merge)
• Context: The current DICE_REGEXP accepts \d*[dD]\d+, which matches bare D0 and D1. Neither is a real die (a 0-sided die is undefined and a 1-sided die has no roll variance), but both occur as map/door labels in published sourcebooks, producing a custom20.png icon where the original label should be (issue (Minor) Replacement of D1 false positive #1126 screenshot).

What changed?

• src/common/utils.js: tightened the bare-die alternative in DICE_REGEXP from \d*[dD]\d+ to \d+[dD]\d+|[dD](?:[2-9]|\d{2,}). Applied in both the primary dice group and the trailing-extra-dice modifier group so compound expressions like 1d8+1d6+3 still parse. Comment added inline explaining the rationale.
• test/dice_regex.test.js (new): 25-case standalone Node script. Loads DICE_REGEXP directly from src/common/utils.js at runtime (no bundler needed) and exits non-zero on any failure.
• package.json: added "test": "node test/dice_regex.test.js".

How to test

1. git fetch && git checkout fix/d1-d0-false-positive-1126
2. npm test — should print 25/25 passed.
3. Manual: load the unpacked extension build (npm run build:chrome / :firefox), open the Vecna ritual chamber page, scroll to the doors sentence — D1 should render as text, not a Beyond20 icon. Sanity check a known-good page (e.g. any monster stat block with 1d20 / 2d6+3) and confirm dice buttons still appear.

Reviewer notes

• Scope is intentionally narrow: only D0 and D1 are excluded. The original issue thread noted D2 also collides with map labels, but d2 is a legitimate (if uncommon) die, so excluding it would regress real dice notation. This PR fixes the impossible-die cases and leaves the D2+ ambiguity as a known trade-off (worth a follow-up if a context-aware approach is wanted).
• Test framework choice: I deliberately did not pull in Jest/Mocha — the test file is a single ~150-line Node script with zero new dependencies. Happy to migrate it to whatever you prefer if you'd rather grow this in a different direction.
• AI use: disclosed above; commit body also notes it. I'm the human responsible for the change and have verified the behavior.

[0xguy07]

@dmportella I ensured I used the correct template this time. I just want to assist in fixing some of the issues which are on the list. We use this extension a lot during our games, and I want to ensure I can help in the ways that I can regarding ensuring that I contribute.