Restate each function's wire from the codecs its classes state

[estebanzimanyi]

Each function's network and wire are stated again once the codec pass
settles every class (parser/enrich.py restate_wire, run right after
state_type_encodings), through the same assess that first states them,
and the exposable count with them. A serialized value reads and writes
through its class's in and out: a Temporal parameter reads through
temporal_from_hexwkb, whose WKB carries the subtype, and a Temporal
result writes through temporal_as_mfjson with its trailing inputs.

Why. The wire was stated before the codec pass and kept the per-class
pick that pass replaces: every Temporal parameter read through
tbigint_in, every Set through bigintset_in. A tfloat read so fails:
tbigint_in('[1.5@2026-01-01 00:00:00+00, 2.5@2026-01-02 00:00:00+00]')
raises 'invalid input syntax for type bigint: "1.5"'.

Measured. Over MobilityDB 72abdf86c2, 4194 wire values named a codec
their class does not state and none does now: 1821 Temporal parameters
from tbigint_in and 689 results from temporal_out, 461 Set, 435 Span and
296 SpanSet parameters from the bigint readers and 492 of their results
from *_out; each parameter now reads through its class's hex-WKB reader,
each Temporal result writes through temporal_as_mfjson and each Set,
Span and SpanSet result through its hex-WKB writer. The six
Raster functions writing a raster become exposable through
raster_as_hexwkb, 3228 to 3234, and none stops being exposable. The same
tfloat, written as hex WKB, reads through temporal_from_hexwkb and writes
back through temporal_as_mfjson as
{"type":"MovingFloat","values":[1.5,2.5],...}.

Witness. tests/test_enrich.py restates a parameter and a result through
a settled codec and the exposable count with them, and over the catalog
finds every wire value naming its class's codec; both fail on the code
without this change. The suite floor goes from 395 to 398.