feat(cache): support manifest file cache

[gripleaf]

Purpose

Add an optional process-local manifest file bytes cache to reduce repeated manifest file reads during scan/planning paths.

This change:

• Adds manifest.cache-max-memory-size, defaulting to 64MB; 0 disables the cache.
• Reuses the existing CacheManager to cache raw manifest file bytes.
• Wires the shared manifest cache manager through CoreOptions, ManifestList, ManifestFile, and IndexManifestFile.
• Reads cached manifest bytes through MemorySegmentInputStream, which keeps the underlying MemorySegment alive without copying bytes.
• Uses a whole-file cache key to avoid extra file-status lookups on cache hits.
• Bypasses the cache when IOHook is active so fault-injection tests still exercise real IO.

Tests

• git diff --check HEAD~1 HEAD
• cmake -S . -B /tmp/paimon-cpp-manifest-cache-build -DPAIMON_BUILD_TESTS=ON -DPAIMON_DEPENDENCY_SOURCE=AUTO ...

Added/updated UT coverage:

• CoreOptionsTest
  • verifies default manifest cache size is 64MB
  • verifies setting manifest.cache-max-memory-size=0 disables the cache
• ManifestFileTest.TestReadUsesManifestCache
  • verifies repeated reads hit cache and avoid repeated file opens
  • verifies filtering still works after reading from cached bytes
• ManifestFileTest.TestManifestCacheCanBeDisabled
  • verifies disabled cache preserves original repeated-open behavior
• ManifestFileTest.TestManifestCacheBypassesWhenIOHookActive
  • verifies cache is bypassed while IOHook is active

Note: local paimon-core-test build was blocked by third-party dependency downloads because the configured proxy could not connect to GitHub.

API and Format

This change adds a new public table option:

• manifest.cache-max-memory-size

It does not change storage format or protocol compatibility. The cache stores raw manifest file bytes in process memory only.

Documentation

Yes. Added docs/source/user_guide/manifest_cache.rst and linked it from the user guide.

Generative AI tooling

Generated-by: OpenAI Codex

[CLAassistant]

All committers have signed the CLA.