[phase-31 2/4] Compaction metadata types

[g-talbot]

Summary

Extend MetricsSplitMetadata with 6 compaction fields and update PostgreSQL model types (Phase 31 Metadata Foundation, PR 2 of 4).

Stacks on gtt/phase-31-sort-schema (PR #6242).

What's included

split/metadata.rs:

• 6 new fields on MetricsSplitMetadata: window_start, window_duration_secs, sort_fields, num_merge_ops, row_keys_proto, zonemap_regexes
• parquet_files: Vec<String> field (replaces single-file assumption)
• Builder methods for all new fields
• TW-1 and TW-2 debug_assert! invariants from ADR-003
• Backward-compat serde defaults for pre-Phase-31 JSON

split/postgres.rs:

• Compaction columns added to InsertableMetricsSplit and PgMetricsSplit
• from_metadata() populates new SQL columns
• SS-5 debug assertions for JSON/SQL column consistency

schema/fields.rs:

• ParquetField::from_name() for name-based field lookup (used by sort field resolver)

Verification

• cargo build -p quickwit-parquet-engine ✅
• cargo test -p quickwit-parquet-engine -- split::metadata split::postgres ✅ (11 tests)
• cargo clippy -p quickwit-parquet-engine --all-features --tests ✅

Test plan

• Pre-Phase-31 JSON backward compatibility test
• Round-trip test with all compaction fields
• skip_serializing_if test for optional fields
• Postgres model round-trip test with compaction columns
• Clippy clean

(Replaces #6243 which was accidentally auto-merged during a rebase)

🤖 Generated with Claude Code

[mattmkim]

@codex review

[mattmkim]

LGTM, let's see if codex finds anything

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b6eb5958d7

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Add serde default for parquet_file in legacy metadata

MetricsSplitMetadataSerde now requires parquet_file, but metadata JSON produced before this commit did not contain that field, so deserialization will fail for existing rows during upgrades. PgMetricsSplit::to_metadata() parses split_metadata_json, and metastore.rs currently drops rows on parse failure via to_metadata().ok()?, which makes legacy splits disappear from list/query results instead of remaining readable.

Useful? React with 👍 / 👎.

[mattmkim]

i mean, true if we were in production already, but i don't think this matters.

[chatgpt-codex-connector]

Derive parquet_file when builder callers omit it

The builder stores parquet_file as a default-initialized String and build() writes it through unchanged, so callers that do not explicitly call .parquet_file(...) now emit metadata with an empty parquet path. This silently records invalid split metadata for current construction paths and loses the file location that was previously derivable from split_id.

Useful? React with 👍 / 👎.