Optionally keep the binary a sample was submitted as (#95)

[r0ny123]

Closes #95 (option to store submitted binaries in MCRIT).

Changes

• STORAGE_KEEP_SUBMITTED_BINARIES (default off): when set, Worker.addBinarySample stores the raw bytes after the sample is indexed and reports binary_stored in the job result. Nothing changes for SMDA report submissions, which carry no binary.
• Storage: storeSampleBinary(), getSampleBinary(), deleteSampleBinary() on the interface. The MongoDB backend keeps them in a GridFS bucket sample_binaries of the storage database, keyed by metadata.sample_id (indexed), replaces on re-store, removes them with deleteSample and drops the bucket with clearStorage. The memory backend keeps them in a dict.
• GET /samples/{id}/binary answers the bytes as application/octet-stream, 404 for an unknown sample or one without a stored binary; McritClient.getSampleBinary() answers bytes or None.

Verification

• tests/testSampleBinaries.py (4 tests): the route serves octets and 404s both ways; the client answers bytes or None; the worker stores the binary for the indexed sample only when configured.
• tests/testStorage.py (both backends): store, read back, replace, delete, and the binary goes with its sample.
• Full suite: 209 passed; ruff check, ruff format --check, ty check clean.
• Live verification on the MongoDB-backed instance follows in a comment.

[r0ny123]

Live verification on the MongoDB-backed instance with STORAGE_KEEP_SUBMITTED_BINARIES = True (server and worker restarted on this change):

• Submitted /usr/bin/cat (39,384 bytes) through McritClient.addBinarySample. The job result carried binary_stored: True for the new sample, and one file sat in the sample_binaries bucket for it.
• getSampleBinary(sample_id) answered bytes equal to the submitted file.
• getSampleBinary(0), a sample that was submitted as an SMDA report, answered None (404).
• deleteSample(sample_id) removed the sample and its bucket file with it (0 files afterwards).