feat: add Parquet merge policy for compaction (Phase 2)

Adds a constant write amplification merge policy for Parquet splits,
adapted from the existing ConstWriteAmplificationMergePolicy but using
byte size instead of document count as the primary size metric. This is
Phase 2 of the Parquet compaction project — the decision layer that
determines which splits to merge within each compaction scope.

Key components:
- ParquetMergePolicy trait mirroring the MergePolicy interface
- CompactionScope grouping by (index_uid, sort_fields, window_start)
- ConstWriteAmplificationParquetMergePolicy with bounded write amp
- finalize_operations() for cold window compaction
- 33 tests: unit, proptest (MC-CONSERVE/LEVEL/WA/IDEMPOTENT), simulation

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: g-talbot

quickwit/Cargo.lock

@@ -30,6 +30,7 @@ ulid = { workspace = true }
 
 [dev-dependencies]
 proptest = { workspace = true }
+rand = { workspace = true }
 regex = { workspace = true }
 tempfile = { workspace = true }
 

quickwit/quickwit-parquet-engine/Cargo.toml

@@ -21,6 +21,7 @@
 //! file has non-overlapping key ranges.
 
 mod merge_order;
+pub mod policy;
 mod schema;
 mod writer;