fix(framework): normalize Unicode command routing

[iriswolf]

Summary

• normalize exact command patterns and incoming command tokens to Unicode NFC
• normalize command-template patterns and bodies while preserving raw Message.Text
• keep CommandRegex and all text route kinds free from implicit normalization
• document that string values bound by CommandTemplate use NFC
• cover direct and generated registration with canonical-equivalence regression tests

Contract

Canonically equivalent sequences now match the same exact or template command. For example, composed ё (U+0451) matches decomposed е plus combining diaeresis (U+0435 U+0308).

This does not alias the distinct letters е and ё. Applications that accept both spellings still declare both routes explicitly.

The original Telegram message text remains unchanged in MessageContext. Regex routes preserve application-owned matching and capture semantics.

Touched hot paths

• TelegramRouteDescriptor normalizes exact/template patterns once during route-table construction.
• TelegramHandlerSelector normalizes only exact command tokens and command-template bodies immediately before matching.
• Already normalized strings are returned unchanged, avoiding replacement-string allocation in the normal path.
• Prefix extraction, bot-mention matching, regex routes, text routes, filter evaluation, and handler invocation are intentionally unchanged.

Validation

• dotnet test .\TeleFlow.sln --no-restore --verbosity minimal — 849 passed
• dotnet format style .\TeleFlow.sln --verify-no-changes --no-restore --verbosity minimal
• .\eng\verify-strict-analyzers.ps1 -Configuration Release -NoRestore
• git diff --check origin/main...HEAD

Documentation

Updated the English and Russian routing fundamentals/reference pages and the unreleased changelog.

Closes #147