add sql formatter and fix parser gaps

[emrberk]

Formatter

Adds format(sql, options) — a token-based SQL formatter, exported from the root and from @questdb/sql-parser/formatter (lexer only, no parser in the bundle).

• Reuses the parser's token definitions through a second, lossless lexer. Whitespace and comments are kept as trivia, unknown characters become opaque tokens, unterminated strings and comments are tolerated.
• Never throws on SQL content. Every token, comment and unknown character is preserved; only whitespace between tokens changes. Unbalanced or unterminated input is preserved verbatim from the point of the problem.
• Layout is driven by a small phrase table per statement kind (SELECT, INSERT, UPDATE, CREATE TABLE/MATERIALIZED VIEW/LIVE VIEW, ALTER TABLE/MATERIALIZED VIEW). Unknown statements get normalized spacing and no line breaks. One width rule (maxLineWidth, default 80): lists, predicates, CASE, IN/VALUES lists, OVER (...), named windows, join sub-clauses and PIVOT (...) stay inline when they fit, otherwise break one item per line.
• Options: indent (default two spaces), maxLineWidth.
• Tests: 63 exact-output fixtures; the full docs corpus and a seeded malformed-input corpus (truncations, delimiter edits, injected garbage) checked against four oracles — identical token stream, identical operator adjacency, idempotence, and identical AST where the parser accepts the input; performance budget for a 5,000-line script and 50-level nesting.

Playground: https://questdb-sql-formatter-playground.netlify.app/

Parser

Gaps found by running the current documentation through the parser and the round-trip test:

• ALTER USER | GROUP | SERVICE ACCOUNT … SET MEMORY LIMIT <size> | UNLIMITED
• RIGHT [OUTER] JOIN, FULL [OUTER] JOIN
• UNNEST(...) after a join, e.g. CROSS JOIN UNNEST(t.asks[2]) u(vol)
• FILL(PREV(other_column), PREV)
• CONVERT PARTITION TO PARQUET … WITH (bloom_filter_columns = …, bloom_filter_fpp = …)
• CREATE LIVE VIEW … OWNED BY owner
• FROM t alias TIMESTAMP(col) (alias before the designation, as the server parses it)
• COPY … TO … WITH … BLOOM_FILTER_COLUMNS … BLOOM_FILTER_FPP …
• ALTER MATERIALIZED VIEW … ADD INDEX TYPE POSTING [DELTA | EF] [INCLUDE (…)] [CAPACITY n]

Serializer: CREATE LIVE VIEW … AS SELECT no longer gains parentheses on output; CREATE TABLE writes FORMAT after WAL.

Corpus

docs-queries.json grows from 1,797 to 2,637 queries: every questdb-sql block in the documentation that was not yet covered (syntax templates, fragments and intentionally invalid examples excluded), plus the queries from the 0.1.16 verification set. All pass parse, round-trip, autocomplete walkthrough and the formatter oracles.

No version bump.