perf(write): write input changelog and data files in parallel

[lszskye]

Purpose

Linked issue: #174

Improve INPUT changelog write performance by writing data and changelog
files concurrently.

This change:

• writes changelog files asynchronously while data files are written on
  the caller thread;
• waits for both write pipelines before publishing their results;
• aborts both writers when either pipeline fails;
• adds changelog-producer.input.parallel-write, enabled by default;
• preserves the previous sequential behavior when the option is disabled;
• extracts common sort-merge writing logic;
• reuses a shared helper when creating data and changelog rolling writers.

Tests

• Extended CoreOptionsTest to verify the default and configured values
  of the new option.
• Updated MergeTreeWriterTest.TestInputChangelog to support
  nondeterministic file numbering during parallel writes.
• Added MergeTreeWriterTest.TestInputChangelogWithParallelWriteDisabled
  to verify the sequential fallback and file creation order.

API and Format

This change adds the public option
changelog-producer.input.parallel-write.

It does not change the storage format or protocol.

Generative AI tooling

Generated-by: OpenAI Codex (GPT-5)

[lxy-9602]

+1