Increase AttributeStore lookup cache size

[Symmetricity]

This PR is AI generated.

Summary

This is the AttributeStore cache-size part split out from #912, as requested
after #913 extracted the vector-layer metadata update cache.

It increases the small thread-local AttributeStore lookup caches from 64 to 256
entries:

• cachedAttributePairPointers
• cachedAttributePairIndexes
• cachedAttributeSetPointers
• cachedAttributeSetIndexes

The lookup behavior, shard selection, locking behavior, and generated output
semantics are otherwise unchanged.

Background

AttributeStore already uses small thread-local direct-mapped caches to avoid
falling through to the shared sharded stores for repeated attribute pair/set
lookups. Large OpenMapTiles runs can have enough hot attribute combinations
that the current 64-entry caches collide frequently.

This keeps the existing cache design, but gives common attribute pairs and sets
more room to remain in the local worker cache.

Performance

During candidate acceptance, this was measured on a warmed Austria
OpenMapTiles run with PMTiles output, no --store, and --threads 8. The
comparison was run in both forward and reverse alternating order on the
submitted/open PR stack used for performance investigations.

Combined six-run result:

metric	baseline mean	candidate mean	delta	delta %
wall time	36.398 s	36.263 s	-0.135 s	-0.37%
user CPU	223.018 s	222.582 s	-0.437 s	-0.20%
system CPU	22.345 s	22.220 s	-0.125 s	-0.56%
peak RSS	4,453,069 KiB	4,402,425 KiB	-50,644 KiB	-1.14%
minor faults	1,198,395	1,168,575	-29,820	-2.49%
voluntary context switches	175,208	170,864	-4,345	-2.48%
involuntary context switches	2,633	2,925	+292	+11.08%

The runtime effect is small, but it reduced minor faults, voluntary context
switches, and peak RSS on that fixture.

Possible Regressions

The intended output behavior is unchanged.

The cost is slightly larger thread-local vectors per worker thread:

• pair pointer cache: 64 -> 256 entries
• pair index cache: 64 -> 256 entries
• set pointer cache: 64 -> 256 entries
• set index cache: 64 -> 256 entries

This is still a small fixed-size per-thread cache and does not change shared
store ownership or synchronization.

Testing

Build:

cmake -S . -B local-builds/attribute-store-lookup-cache-size -DCMAKE_BUILD_TYPE=RelWithDebInfo
cmake --build local-builds/attribute-store-lookup-cache-size --parallel 8

Code style:

git diff --check

Semantic checks from candidate acceptance:

fixture: Liechtenstein Geofabrik extract
profile: OpenMapTiles profile

baseline repeat changed_tiles 0
candidate repeat changed_tiles 0
baseline -> candidate changed_tiles 0

[systemed]

Thanks!