KAFKA-20538: Make RocksDBTimeOrderedKeyValueBuffer headers-aware

[aliehsaeedii]

Problem: RocksDBTimeOrderedKeyValueBuffer (the persistent buffer used by stream-table joins with grace periods) was missed when headers were propagated through Streams serdes. Two bugs:

1. evictWhile() deserializes with the wrong headers. Both key and value deserializers are called with internalContext.headers() — the headers of whatever record the processor is currently handling, not the record being evicted. Since records can sit in the buffer for the entire grace period, these almost never match.
2. put() does not persist the record's headers. It stores the caller-supplied recordContext directly in BufferValue, so record.headers() is never copied into the persisted ProcessorRecordContext and is unrecoverable at eviction time.

This PR fixes the followings:

• put() builds a fresh ProcessorRecordContext carrying record.headers() and stores that in BufferValue, so the headers are persisted with the record
• evictWhile() uses bufferValue.context().headers() for both key and value deserialization, so each evicted record is deserialized with the headers it was buffered with.

Testing: Unit tests

• New tests assert headers propagate through eviction, that the record's headers (not the surrounding processor context's) are what surface, and that mutating the live processor context between put and evict does not affect the evicted record's headers.

[github-actions]

A label of 'needs-attention' was automatically added to this PR in order to raise the
attention of the committers. Once this issue has been triaged, the triage label
should be removed to prevent this automation from happening again.