feat: add SERIAL, BIGSERIAL, UUID, and TIMESTAMP WITH TIME ZONE columns

[omer-cengel]

Summary

Schema snapshots can now declare SERIAL, BIGSERIAL, UUID, and
TIMESTAMP WITH TIME ZONE columns, and queries over them generate compiling,
type-safe Java. Previously these spellings failed as unsupported column types,
and TIMESTAMP(n) WITH TIME ZONE was silently reduced to TIMESTAMP because
the schema parser truncated a declared type at its first parenthesis.

Changes

• Canonicalize a declared SQL type by upper-casing it, removing its
  parenthesized arguments, and collapsing whitespace, so
  TIMESTAMP WITH TIME ZONE resolves with or without a precision.
• Accept SERIAL and BIGSERIAL as the delivered integer and bigint mappings,
  and model such a column as not nullable, because PostgreSQL defines them as
  those types with a sequence default and NOT NULL.
• Map UUID to java.util.UUID and TIMESTAMP WITH TIME ZONE to
  java.time.OffsetDateTime in generated method parameters, result record
  components, and imports.
• Reserve UUID and OffsetDateTime as generated type names so a query named
  after one of them cannot collide with the import it needs.
• Cover the new types with schema-parser and generator tests, generated-source
  compilation, and PostgreSQL round-trip tests including null binding and null
  result reading for the nullable uuid and timestamptz columns.

Scope and non-goals

• Only those four spellings are accepted, case-insensitively, plus an optional
  precision on TIMESTAMP WITH TIME ZONE. TIMESTAMPTZ, SERIAL4, SERIAL8,
  SMALLSERIAL, and TIMESTAMP WITHOUT TIME ZONE continue to fail as
  unsupported, so one spelling can never resolve to two mappings.
• No other CREATE TABLE form is added: column defaults, references, and check
  constraints are unchanged.
• Generated APIs keep using reference types; no Optional or nullable wrapper
  is introduced, and the JDBC runtime is unchanged.