docs: document bucket bot_token override

Surface the per-bucket Telegram bot token support that landed in 124c695
inside the default config and README so users can discover it without
reading the config-layer code.

- data/config.yaml: switch telegram block to active bot_token with the
  legacy bot_token_env kept as a commented deprecated fallback; add
  per-bucket bot_token override comments on the example buckets.
- README.md: note the optional per-bucket bot_token in the
  public-files example and trim the env var description to a single
  line.

Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>

feat(store): use non-verifiable ETag for typed uploads

Typed uploads (photo/video/audio/animation) may be recompressed by
Telegram, so the MD5 of original input bytes doesn't match the actual
downloadable content. Use a random ETag with "-typed" suffix (similar
to multipart ETag format) so S3 clients like rclone don't attempt MD5
verification on these objects.

Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>

ci: build docker images by digest

Generated with [Claude Code](https://claude.ai/code)
via [Happy](https://happy.engineering)

Co-Authored-By: Claude <noreply@anthropic.com>
Co-Authored-By: Happy <yesreply@happy.engineering>