Use Case

As a... developer using insertMany/bulkWrite at high volume.
I want... the driver to stop serializing each document twice per write.
So that... bulk writes consume less CPU and stop blocking the event loop.

User Experience

• Bulk writes (insertMany, bulkWrite, fluent bulk ops) do one BSON pass instead of two, cutting encode CPU by roughly half.
• The gain is synchronous, event-loop-blocking work removed, so apps see lower latency on other operations sharing the process during large bulk writes. Biggest wins on high doc counts and many-field documents.
• Not a bug; no correctness impact. Behavior is unchanged (including ignoreUndefined defaults).

Dependencies

• N/A

Risks/Unknowns

• Switches the wire encoding of documents/updates/deletes from a type-0 array to an OP_MSG document sequence (type 1); command monitoring must reconstruct these to arrays or APM assertions break.
• Must keep the array path under auto-encryption (libmongocrypt needs the plaintext command).

Acceptance Criteria

Implementation Requirements

• Serialize each op once in the bulk path; reuse the bytes for the max-size check, batch splitting, and the wire message.
• Preserve batch-splitting behavior and default ignoreUndefined: false semantics (no data-shape change on upgrade).
• Fall back to the array path under auto-encryption and for non-bulk callers.

Testing Requirements

• Existing CRUD / bulk / retryable-write / transactions / CSE spec tests pass unchanged.
• driver_bench small_doc_bulk_insert / large_doc_bulk_insert show improved throughput in CI.

Documentation Requirements

• N/A

Follow Up Requirements

• N/A