feat(usage): report per-model plan limits in their own bar row

[serialexperimentslainnnn]

What

• Fable usage is reported. Per-model windows live in rate_limits.model_scoped — an array the parser walked past. Reading it is necessary and not sufficient: the binary synthesises that key behind its own remote config (IUt(limits, jJe()) returns [] when tengu_usage_overage_included_models is empty, and the key is spliced in only when the projection yielded something), so a --print session never received it. The raw rate_limits.limits[] array it projects from does arrive untouched, so weekly_scoped entries naming a model are read from there too — the binary's filter, without its allowlist, since that list selects overage billing, not which limits meter you. resets_at is epoch seconds there as often as a string, so it is normalised rather than deserialized (a numeric one would have dropped the window in silence).
• No more flicker. loadPlanRateLimits gives the usage endpoint 5 s and falls back to seedUtilization(), an object rebuilt from the rate-limit response headers that structurally carries only five_hour/seven_day, flagged "seeded" and accepted identically downstream. A failed poll was therefore indistinguishable from one saying the per-model window is gone. A refresh is now merged into the last one by window key.
• The plan limits get their own responsive row under the status line, one labelled bar per window, instead of dots trailing a wrapping row of unrelated metrics.
• Also in this cycle: the "Other models" picker group (with set_model as a correlated request, so a refusal restores the previous model), nimbus_quill hidden on both ingestion paths, and quota notifications titled through UsageWindow.title(key).

Evidence

Every claim about the binary was read out of claude 2.1.223, not inferred, and confirmed live against idea.log:

usage window model_scoped:Fable: utilization=17.0 -> 17%
get_usage raw limits[]: [… {"kind":"weekly_scoped","percent":17,
  "scope":{"model":{"display_name":"Fable"}}, …}]

(that diagnostic log was temporary and is not in this branch)

Gates — run clean locally (clean … --rerun-tasks)

Gate	Result
test	780 tests, 0 failures
koverVerify / detekt / spotlessCheck	pass
verifyPlugin	Compatible: IC-251, IC-252, IU-253, IU-261, IU-262
buildPlugin	0 node_modules entries, META-INF/{LICENSE,THIRD-PARTY-NOTICES.md} present
npm test / lint / format	111/111, clean
npm audit --omit=dev	0 vulnerabilities

Manually validated in a real IDE (PyCharm 2026.2) before release.

🤖 Generated with Claude Code