[YuangGao]

Which issue does this PR close?

• Follow-up to fix!(io): bound S3 write request size with a configurable part size #3179.

What changes are included in this PR?

Addresses the review nits on #3179:

• Note why the part-size cast can't fail for parsed values.
• Document why default_multipart_part_size is a named serde default.
• upload_part_count returns None for a bare-MD5 ETag, so the single-request test asserts it went out as one request.

Are these changes tested?

• The S3 tests pass against the RustFS fixture.

AI Disclosure

• AI-assisted implementation.