Parser: fix exponential parse time on speculative prefix parsing

[LucaCappelletti94]

parse_prefix speculatively interprets the next token as a reserved-word expression head and, on failure, falls back to treating it as an unreserved identifier. Both arms can independently recurse into the same downstream position, so inputs like IF(current_time(current_time(...x were re-walked at every level for 2^N total work. Each parse_prefix failure is now memoized by start position so the second visit short-circuits.

Same family as #2344 and the speculative NOT prefix fix on pathological-combined. PostgreSQL, release build:

depth	size	before	after
10	134 B	900ms	399us
15	199 B	>60s	110us
20	264 B	>60s	109us
25	329 B	>60s	126us
30	394 B	>60s	124us

Regression test in tests/sqlparser_common.rs runs the parse on a worker thread with a 5-second hang guard. Criterion bench under sqlparser_bench tracks perf at n=10/20/30.

[LucaCappelletti94]

Me watching yet another fuzzer crash: