fix: keep Russian root match from absorbing conjunctions

[alyldas]

Summary

• Keep the Russian huy base rule from treating a following standalone conjunction as part of the same loose stretched root.
• Replace the final root character class with explicit base variants so loose stretching repeats the selected variant rather than the whole variant set.
• Add regression coverage for match ranges, censor output, and Telegram HTML spoiler rendering.

Validation

• npm run check with a temporary npm cache
• npm run benchmark:profanity on origin/main
• npm run benchmark:profanity on this branch

Benchmark Evidence

Baseline origin/main:

• createProfanityFilter(): 7.6038 avg ms
• check short clean: 0.0275 avg ms
• check cyrillic clean: 0.0615 avg ms
• check long clean: 4.5850 avg ms
• check loose match: 0.0229 avg ms
• analyze short match: 0.0309 avg ms
• censor short match: 0.0356 avg ms

This branch:

• createProfanityFilter(): 7.7734 avg ms
• check short clean: 0.0263 avg ms
• check cyrillic clean: 0.0611 avg ms
• check long clean: 4.3767 avg ms
• check loose match: 0.0218 avg ms
• analyze short match: 0.0315 avg ms
• censor short match: 0.0325 avg ms

Runtime behavior changes only for the Russian base root boundary case where a following standalone conjunction was previously absorbed by loose stretching.

Compatibility Notes

• Public APIs, exports, scanner contracts, and package metadata are unchanged.
• Existing Russian loose stretching remains available for the explicit base variants while the хуй и range now ends after the obscene token.
• No workflow, permission, secret, or packaging compatibility changes are introduced.

Closes #144

No publish, no merge, no tag/release.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4956df1f95

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Restore reviewed хуей/ху-ей loose matches

With this source split into fixed alternatives, stretch: true can only repeat the selected final letter, so inputs like хуей or ху-ей now only produce a хуе prefix match; the loose boundary check then rejects it because the same token still has a trailing й. These are existing reviewed loose corpus cases in tests/loose-corpus.spec.ts that should be fully masked, so this change creates a bypass while fixing the following-conjunction case.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Stop absorbing same-letter following words

Because each explicit alternative is still stretched, a following standalone word that starts with the same final letter is consumed as another repeat of that final atom. For example, привет хуи и мир still gets a loose range over хуи и rather than just хуи (and хуя я ... has the same shape), so the range leak this patch is meant to fix remains for base variants whose last letter matches the next word.

Useful? React with 👍 / 👎.