Say which documents are over MongoDB's size limit, and keep the sample (#42)

[r0ny123]

Closes #42 (Mongo: BSON document might be bigger than 16 MB).

Cause

An insert batch with one document over MongoDB's 16 MiB document limit failed as a bare ValueError("Database insert failed."), with the error log carrying only the traceback, so nothing said which sample or function was responsible, and the sample was lost. The reporter saw it 4 times in 120k files. Since v1.7.0 the disassembly lives in its own xcfg documents, one per function, which is where a single giant function ends up over the limit.

Two shapes of the error exist: pymongo raises DocumentTooLarge before sending a clearly oversized document, and MongoDB answers a write error (code 2, "object to insert too large") for one just over the limit, which pymongo surfaces as BulkWriteError after inserting the documents before it (ordered insert).

Fix

_dbInsertMany recognises both, measures the documents the ordered insert did not get to, and logs the oversized ones with their identifying fields (_id, function_id, sample_id, sha256, offset) and byte sizes.

• For the blob collections (xcfg, query_xcfg) it drops the oversized blobs with a warning and inserts the rest. The sample and all its functions are stored; only the disassembly of the affected function is missing, which the readers already treat as "no disassembly" (they decode a missing blob to {}). That function then gets no MinHash, as any function without disassembly.
• For any other collection it raises a ValueError naming the documents, since a function or sample document that cannot be stored cannot be dropped.

Verification

• tests/testStorage.py (Mongo): a blob one byte over the limit is dropped with a warning naming one function, the other blob is stored, and the error log records the offender with its size and that it was dropped; an oversized blob in the middle of five keeps the other three; an oversized function document fails with a message naming exactly that document and the ordered insert's earlier document stays stored.
• Storage suite: 52 passed; ruff check, ruff format --check, ty check clean.
• The tests run against a real MongoDB 7 and exercise the server-side write error path (the one the issue's 60 MB case would take through pymongo's own check).