fix(postgres): add missing TO_CHAR format tokens (Mon, Month, Day, Dy, AM/PM, HH)

[mwade0118]

Summary

Adds bare (non-TM-prefixed) PostgreSQL TO_CHAR format tokens to Postgres.TIME_MAPPING:

• Month names: Mon/mon/MON → %b, Month/month/MONTH → %B
• Weekday names: Day/day/DAY → %A, Dy/dy/DY → %a
• AM/PM: AM/PM/am/pm/A.M./P.M./a.m./p.m. → %p
• Bare HH: HH/hh → %I (PostgreSQL defaults to 12-hour when no suffix)

Also adds INVERSE_TIME_MAPPING to prefer canonical bare forms (Mon, Day, Dy, Month, HH12, AM) when generating Postgres SQL.

Problem

Previously, these tokens were either left as unconverted literals or partially mangled:

PG Token	Expected	Actual (before)	Issue
Mon	%b	Mon (literal)	Not converted
Month	%B	Month (literal)	Not converted
Day	%A	%uay	D → %u matched first, ay left as literal
Dy	%a	%uy	D → %u matched first, y left as literal
AM/PM	%p	literal	Not converted

The TM-prefixed versions (TMDay, TMDy, TMMon, TMMonth) were already mapped. This PR adds the standard bare forms which are far more commonly used in practice.

The trie-based format_time() correctly prioritizes longer matches, so Day → %A takes precedence over D → %u without affecting the existing single-character D mapping.

Note: Redshift already extended Postgres with "MON": "%b" (line 27 of redshift.py), which is now redundant but harmless.

Fixes #7476

Test plan

• Added 7 new validate_all tests in test_postgres.py covering all new format tokens
• Updated 2 existing identity tests to account for HH/HH12 and PM/AM canonicalization
• Updated 1 Exasol test expectation (TMDy → Dy for postgres write target)
• All 862 existing tests pass (804 dialect + core tests, 58 bigquery tests)