Parser: fix exponential parse time on compound chains

[LucaCappelletti94]

parse_compound_expr recursed into parse_subexpr after every ., which re-walks the rest of the chain inside a rollback boundary. On inputs like IF a.b.c...x.# the work doubled per chain element, giving 2^N parse times. Switching the inner call to parse_prefix removes the redundant traversal: the outer loop already walks the chain, so the resulting AST is identical on valid SQL.

Measured on PostgreSqlDialect with three inputs from a libFuzzer corpus, release build:

Input	Before	After
iF i.D.i.:Fi.... (65 B)	419 ms	87 us
iF i.D.i.i. ... .*~ (58 B)	893 ms	69 us
if-stf-localtclocal33alt.... (306 B)	>35 s	98 us

Regression test in tests/sqlparser_postgres.rs runs a 30-element chain with a 5 s timeout. Pre-fix it hits the timeout, post-fix it finishes in well under a millisecond.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.