feat(bindings): tbox WKB/hex-WKB serialization + hashing

[estebanzimanyi]

Closes the user-facing tail of `temporal/021_tbox` (was 80%, brings the user-visible surface to ~100%).

New SQL surface

Signature	Routed through
`asBinary(tbox)` -> `bytea`	MEOS `tbox_as_wkb` (`WKB_EXTENDED`)
`asHexWKB(tbox)` -> `text`	MEOS `tbox_as_hexwkb`
`tboxFromBinary(bytea)` -> `tbox`	MEOS `tbox_from_wkb`
`tboxFromHexWKB(text)` -> `tbox`	MEOS `tbox_from_hexwkb`
`tbox_hash(tbox)` -> `integer`	MEOS `tbox_hash`
`tbox_hash_extended(tbox, bigint)` -> `bigint`	MEOS `tbox_hash_extended`

What stays unregistered (and why)

The remaining `021_tbox` names that the audit (PR #66) flags as missing:

• `tbox_in`, `tbox_out`, `tbox_recv`, `tbox_send` — PostgreSQL I/O protocol internals; DuckDB exposes `tbox` via `CAST` to/from VARCHAR which is already wired.
• `tbox_extent_combinefn`, `tbox_extent_transfn` — aggregate transition functions; live in the separate aggregate-infrastructure track (PR test(parity): 015_span_aggfuncs.test — aggregate surface manifest #21).
• `tnumber_joinsel`, `tnumber_sel` — PostgreSQL planner selectivity hooks; no DuckDB equivalent.

Tests

• `test/sql/parity/021_tbox_wkb_hash.test` — 6 assertions, including binary and hex-WKB round-trips.
• Full suite: 758 assertions / 14 test cases passing under `TZ=UTC`.

Coverage delta

Per the audit in PR #66:

• `temporal/021_tbox`: 48/60 (80%) → 54/60 (90%) by name (the remaining 6 are infra-only).

Test plan

• `cmake --build . --target shell unittest` clean
• New parity test passes (6/6)
• Full suite green (758/14)

[estebanzimanyi]

Superseded by the consolidated PR branch consolidate/io-serialization-parity. All changes from this PR are included in that branch as a single squashed commit. Please review and merge the consolidated branch instead.