Optimize CBOR Codec

[adwsingh]

Issue #, if available:

Description of changes:

• Replace Sink interface with direct byte[] buffer writes + VarHandle for multi-byte big-endian ops — no virtual dispatch per byte
• Pool serializer instances via striped AtomicReferenceArray to avoid repeated buffer allocation
• Pre-compute CBOR-encoded field name bytes at schema load time; write them with a single System.arraycopy
• Inline UTF-8 encoding with an ASCII fast path: deprecated String.getBytes(int,int,byte[],int) bulk-copies, OR-accumulator detects non-ASCII, rewind path encodes UTF-8 directly into the buffer without allocation
• Use a long bitmask instead of boolean[] for tracking indefinite-length collection nesting
• Fuse ensureCapacity for blob writes — single capacity check covers CBOR header + payload, eliminating redundant buffer growth
• Add Codec.serialize() that returns a ByteBuffer directly from the pooled serializer

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[rhernandez35]

I avoided doing this originally because I didn't want adding new fields to have visible performance effects, especially to legacy clients. Adding a new optional field or reordering schema fields should not have a visible performance effect.

[rhernandez35]

Please do some perf testing with clients that have out-of-date schemas and servers that are receiving unknown fields. I'm concerned about adverse impact outside of all but the happiest cases in this PR.

Primary concern addressed, got 1 approval if there are more concerns we can address in a followup.