Update Sentencepiece Tokenizer

[apaniukov]

• Upgraded SentencePiece to 0.2.2 for both Python and C++ builds, including CMake URL/hash, dependency metadata, and lockfile updates.

• Updated C++ integration for SentencePiece 0.2.2: switched removed logging API usage to Abseil logging, adjusted Abseil provider configuration, and hardened precompiled charsmap generation status handling.

• Fixed SentencePiece/HF normalization regressions:

  • CharsmapStep now defaults remove_extra_whitespaces=False.
  • HF Precompiled normalizers preserve whitespace.
  • WhitespaceSplit conversion now matches HF behavior with trim + whitespace collapse + split.

• Fixed SentencePiece runtime errors for tokenizer conversion by preserving BOS/EOS ids when mutating tokenizer-side SentencePiece models.

• Removed pytest-harvest from dev dependencies and replaced README coverage generation with pytest terminalreporter stats, making --update_readme -n auto work reliably under xdist.