fix+feat(kiota): schema-default stripping (run #7's five 400s) + the final wave — full 23-resource parity

[ShocOne]

Why

Two things in two commits: the root-cause fix for acceptance run #7's five failures, and the final wave of the parity plan.

Commit 1 — why five tests_* creates returned bare 400s

Reproduced offline by serializing the generated constructResource output for the dns_server minimal fixture and diffing against the recorded accepted body: ours carried one extra field, "networkMeasurements": true. Kiota constructors stamp every spec-declared default: onto the model; construct overwrites the fields the schema wires (nil when unset), so what leaks is exactly the defaulted-but-unwired fields. Confirmed against the live lab: the body with the leaked default → 400 ("Cloud Agents are not supported for bandwidth measurements" — the API's own words); byte-for-byte without it (= what the recordings sent) → 201 (object deleted immediately after). The five failing resources are precisely the ones whose request models default fields the curation doesn't wire. Response models carry the mirror defect: a constructor default masks field absence from flatten.

Fix, through the document-patch machinery from #109:

• docpatch gains strip-schema-defaults — an op RFC 6902 can't express: every schema in the document loses its default, walking the schema grammar (properties/items/allOf/…) so a property literally named default and a default key inside an example both survive; a document with no defaults refuses as stale. Unit-tested.
• A third ThousandEyes patch applies it. Regenerated constructors now set nothing but their AdditionalData map, and a client wire test pins that a fresh request model serializes only what was set.
• Bonus from the same debugging session: the errors scaffold now quotes RFC 7807 title/detail from the deserialized error model when kiota's message is its say-nothing fallback — the 400 above carried its full explanation while the diagnostic said "The API offered no further explanation".

Commit 2 — the final wave

alert_rule, dashboard, dashboard_filter complete parity: the kiota pilot now binds the same 23 resources, 2 data sources, 1 ephemeral, 1 action, 1 list facet as the resty pilot — 1126 verified bindings. The single-nested-heavy trio ported as pure blueprint data; the one manual touch was dropping resty's named-slice cast on dashboard_filter's context expand (no counterpart in a method-typed SDK).

Verification

• bindings check ✅ 1126/1126; probe verify exit 0 (all recorded subjects reproduce); sdk generate -check ✅ with all three patches; provider drift 0 on both pilots; all suites 0 failures.
• Offline: dns_server constructed body now byte-identical to the recorded accepted request.
• Live spot-check on the lab tenant (with the shared token): leaked-default body 400, stripped body 201, created object deleted.

After merge: dispatch go | Acceptance tests (kiota) — this should be the first fully green run at complete parity. Then the lab token can be cycled.

🤖 Generated with Claude Code