fix(slack): resolve model aliases before LLM fallback

[decofe]

Summary

• resolve common natural-language harness, model, provider, and reasoning aliases before calling the LLM selector
• allow one-character fuzzy matches for longer aliases inside explicit selection clauses
• retain the LLM fallback for ambiguous requests and guard against incidental model-name mentions

Validation

• pnpm --filter slackbotv2 test (252 passed, 1 skipped)
• pnpm --filter slackbotv2 run check:types (blocked by three existing fetch.preconnect errors in untouched test/slack-user.test.ts)

Prompted by: @0xalpharush