[S2] compress.py "lossless guarantee" is false — BoneToken.entry discarded on decode

[erinepshovel-code]

compress.py module docstring states:

Lossless guarantee: pt2 and metrics2 are token-for-token identical to the originals.

_dict_to_tok (the decode path) reconstructs BoneToken objects with entry={}:

BoneToken(..., entry={})   # raw canon dict always empty after decode

Five of the six BoneToken slots round-trip correctly (surface, normalized, bone_type, primary, families). The entry field (the raw canon dictionary for the matched bone) is set to empty dict on every decode.

Impact: Any consumer that reads token.entry after a compress/decompress cycle gets {} instead of the original canon entry. Downstream code that reconstructs features from token.entry silently produces empty or wrong results.

Fix options:

1. Serialize entry into the compressed format and restore it on decode (accepts the size cost).
2. Correct the docstring: "lossless for surface/normalized/bone_type/primary/families; entry must be re-resolved via canon lookup after decode."
3. If entry is intentionally excluded for size, add a note directly on BoneToken that entry is not guaranteed to be populated after decode.