Add bounded multipart upload parallelism

[jiuker]

fix: #216

Add bounded multipart upload parallelism

PutObject now supports concurrent UploadPart calls via the new PutObjectArgs::max_inflight_parts field (default 1 = sequential behavior preserved).

Changes

File	Change
include/miniocpp/args.h	Added unsigned int max_inflight_parts = 1 to PutObjectArgs
src/args.cc	Validate max_inflight_parts > 0
src/client.cc	Refactored PutObject into sequential + parallel branches
tests/tests.cc	Added PutObjectWithInflight() — MD5 integrity test

src/client.cc highlights

• Sequential path (max_inflight_parts <= 1): Original single-buffer loop, unchanged.
• Parallel path (max_inflight_parts > 1):
  • N page-aligned buffers allocated via AlignedBuffer pool.
  • Producer/consumer model: main thread reads stream, std::async dispatches UploadPart.
  • Bounded inflight cap via std::deque + drain_one lambda.
  • FIFO ordering preserved by part number — CompleteMultipartUpload doesn't care about completion order.
• Bug fix: Corrected one_byte read-ahead indexing in the unknown-size path (buf[part_size] instead of buf[part_size + 1]).

Testing

PutObjectWithInflight() generates 10MB of random data, computes the original MD5, uploads with max_inflight_parts = 1, 2, 4, downloads each object, and verifies MD5 matches the original.