v2 translator echoes wildcard sentinels the new value bounds refuse

[hokiepokedad2]

TrackingV2Translator copies v1 integers to the v2 body verbatim. Several of those columns hold a wildcard sentinel that PoracleNG writes itself, emits nothing for on read, and — once jfberry/PoracleNG#217 lands — refuses on write.

The sentinel is coherent on their side: omit the field and the write default stores the sentinel; read the sentinel back and you get null. Omission round-trips. Echoing does not, and echoing is what we do.

PoracleTrackingProxy.PutV2Async maps 422 to AlarmValidationException with no v1 fallback, so this surfaces as a user-facing validation error on edit, not a silent downgrade.

Affected fields

Taken from #217's head — read wildcard (ptrUnless) against the bound the same field now carries.

Type	Field	Stored wildcard	New bound	When it bites
pokemon	size	-1	minimum: 1	any rule not filtering on size — the default
pokemon	rarity	-1	minimum: 1	any rule not filtering on rarity — the default
pokemon	min_iv	-1	minimum: 0	any rule with no IV floor
raid	level	9000	1–90	every rule with a specific pokemon_id — PoracleNG forces level to 9000 there
maxbattle	level	9000	1–90	same
nest	pokemon_id	0	minimum: 1	a nest rule tracking any species

Raid and max battle are the worst of these: the value we echo is one PoracleNG stamped on the row itself, so it isn't a rule anyone typed wrong. level: 90 ("any tier") sits inside the bound and is fine; only the 9000 placeholder fails.

Egg level is already handled — PositiveIntegers in the type spec, added when 5.2.1 shipped minimum: 1 on that one field. This is the same defect, and the other nine types were never checked. Second time on this shape.

Fix

Add a per-type wildcard table to TrackingV2Translator and drop a field from the outbound body when its value equals the sentinel. Dropping is correct rather than merely convenient: v2 PUT is full-replace, and every one of these fields' write default is the sentinel, so omitting stores exactly what was there.

null may work as an explicit clear — the fields are nullable: true and huma should skip the bound on null — but that is unconfirmed upstream, and omission needs no confirmation. Use omission.

Tests

TrackingV2TypeTranslationTests asserts our field tables against 5.2.1's openapi.golden.json, which is what should have caught this and didn't — it checks that every schema field is classified, not that every value we can emit is acceptable. Add a case per row of the table above: a v1 row holding the sentinel must produce a body with that key absent. Watch each one fail first — a test written next to the fix passes whether or not the fix is right.

Then extend the golden assertion to compare our sentinels against the schema's minimum/maximum, so the next bound added upstream fails the build instead of production.

Timing

Not live. The bounds arrive with #217, which is open. We should land the fix before adopting a build that carries it, and pin Poracle:TrackingApiVersion=v1 as the escape hatch if the ordering slips.