CLM v1.9.0

Added

• The slide authoring tooling now works on C#/C++/Java/TypeScript decks, not
  just Python. CLM's build has long been multi-language, but the authoring
  tooling — slides split / unify, sync, voiceover extract / inline,
  assign-ids, normalize, validate, coverage / lang-coverage /
  language-view, deck discovery, and translate / bootstrap — hardcoded the
  Python # comment token and .py extension. The deck's comment token (//
  for the c-family, # for Python) and extension are now threaded through every
  authoring consumer, resolved per file from the extension via
  comment_token_for_path(). The //-family languages (C#, C++, Java,
  TypeScript) get the same split/sync/voiceover authoring workflow Python already
  had, including .de/.en split companions that preserve the deck extension
  (*.de.cs, *.en.cpp). The default everywhere stays #, so Python output
  is byte-identical. This completes the multi-language authoring migration whose
  structural prerequisite (the header-line-less title convention) shipped in
  1.8.4. Investigation + validation:
  docs/claude/multi-language-authoring-tooling-investigation.md §10.

Changed

• sync translate prompts are now language-aware. A code cell in a //-deck
  is translated as that language (e.g. C# is translated as C#, not "runnable
  Python"), and the model is instructed to preserve // comment prefixes. The
  programming language folds into the translation-cache key for non-Python decks,
  so cached Python translations are unaffected.
• --help text and clm info topics use a language-neutral <ext>
  placeholder instead of implying .py-only, and two now-inaccurate
  "Python-only" claims in the info topics were corrected.

Fixed

• The build no longer silently drops voiceover on //-family decks. The
  host-side voiceover companion merge and the authoring voiceover writers were
  Python-token-only, so a voiceover_*.cs / *.cpp companion was never merged
  into the built notebook. The merge, the companion writers, and extract /
  inline are now token-aware, and the SKIP_OUTPUT globs were broadened so
  voiceover_*.{cs,cpp,java,ts} companions never leak into output.
• assign-ids / normalize / validate now discover .cs / .cpp / .java
  / .ts decks. Deck discovery (rglob + is_slides_file) and validate
  kind-inference were Python-extension-only; the //-family decks are now found
  and dispatched.
• Comment-prefix stripping removes the literal prefix, not a character set.
  _strip_comment_prefix previously used lstrip("// "), which strips any
  leading / or space — turning // /usr/bin into usr/bin and collapsing a C#
  /// doc comment. It now strips the exact // prefix once.