feat(timeseriesql): InfluxQL compat overhaul - parser rewrite, query fixes, DDL hardening

[austin-barrington]

Description

Rewrites the InfluxQL query parser and translator for broad InfluxQL compatibility, hardens the DDL parser against silently-dropped clauses, and fixes correctness issues across the SQL generation layer.

Parser/lexer rewrite - New masking scanner (scan_chars) replaces the ad-hoc state tracking in split_clauses, split_top_level_commas, find_top_level_operator, and split_statements. This fixes panics on non-ASCII characters, semicolons inside regex literals, quoted identifiers containing keywords, and regex-vs-division ambiguity. Also adds OR/AND precedence (OR binds looser), left-associative arithmetic, bare DISTINCT keyword, duplicate clause detection, modulo, bitwise rejection, ::field/::tag cast validation, and source-spelling preservation for keyword-named identifiers.

Query translator fixes - fill() grid correctness (TO anchor extends one step, fill(N) emits INTERPOLATE, fill(null) is default, SELECT INTO never fills), ORDER BY DESC+fill subquery wrapping, tz() flowing into bucket/group-by/fill-anchor expressions for DST-correct daily bucketing, per-series LIMIT/OFFSET via LIMIT n BY, raw selects with GROUP BY tag projecting without SQL GROUP BY, per-point transforms (difference/moving_average/elapsed) without GROUP BY time, moving_average full-window gating, elapsed toNullable defaulting, tag-vs-numeric constant-false, count(distinct)/distinct correct translation, median/percentile/stddev/mode matching InfluxQL semantics, materialized views using LEFT JOIN, and precise __timeâ��time renaming.

DDL parser hardening - CREATE USER parses both WITH PASSWORD/WITH ALL PRIVILEGES orderings, negative durations rejected (zero normalizes to infinite), SHOW clauses accept ON/FROM in either order, SHOW MEASUREMENTS WITH MEASUREMENT rejected loudly, trailing tokens rejected, CREATE DATABASE duration validation, REPLICATION range check, and CREATE MATERIALIZED VIEW AS SELECT surviving the trailing-token check.

Digest normalization - Identifiers keep original case (InfluxQL is case-sensitive), binary expressions are parenthesized for structural distinction, and ON/LIMIT/OFFSET/tz() are included.

Point coalesce - group_and_coalesce_by_measurement replaces coalesce_points_within_measurements; zero-merge case borrows via Cow instead of cloning.

Duration safety - to_nanos uses saturating_mul.

Related issues

Checklist

• cargo fmt --all --check passes
• cargo clippy --all-targets -- -D warnings passes
• cargo test --lib passes
• cargo test --test '*' passes
• New features have documentation (user-guide or developer-guide)
• New config keys have entries in docs/user-guide/configuration.md and config.toml.example
• New metrics follow the naming convention (hyperbytedb_*)
• Breaking changes are noted in the PR description

Release notes

Changed - Query parser and translator rewritten for broad InfluxQL compatibility: correct operator precedence, fill() grid semantics, tz()-aware bucketing, per-series LIMIT, per-point transforms, and aggregate function parity (median, percentile, stddev, mode, distinct, count(distinct)). DDL parser now rejects silently-dropped clauses (negative durations, SHOW MEASUREMENTS WITH MEASUREMENT, trailing tokens, CREATE USER ordering). Duration arithmetic saturates instead of overflowing.

Test plan

~60 new unit tests in parser.rs, to_clickhouse.rs, ddl_parser.rs, ast.rs, digest.rs, and lexer.rs covering masking scanner edge cases, operator precedence, fill grid correctness, tz/DST bucketing, LIMIT BY, raw selects, per-point transforms, tag-numeric comparisons, non-ASCII handling, and DDL parsing. New integration tests in combination_tests.rs for median, stddev, mode, distinct, count_distinct, fill grid (leading/trailing gaps, DESC, defaults, offset), LIMIT per-series, raw select with GROUP BY tag, per-point transforms without GROUP BY time, tag-numeric returning empty, and DST-correct daily bucketing. Existing tests updated to match corrected translations.