docs(api): deprecate the retired app-control routes, correct the appstats payload

[MorningLightMountain713]

Five endpoints changed in FluxOS. This spec is the contract third parties build against, so it is wrong on the day the release lands.

What changed

Endpoint	Now	In the spec
/apps/apppause	in-band 410, use appstop	deprecated: true, real error body as the example
/apps/appunpause	in-band 410, use appstart	same
/apps/startmonitoring	in-band 410, no effect	same
/apps/stopmonitoring	in-band 410, no effect	same
/apps/appmonitorstream	route deleted — plain 404	documented as 404, points at /apps/appmonitor
/apps/appstats	payload narrowed	example replaced with the real shape

The four in-band ones are marked deprecated rather than deleted, because they still answer and a caller looking one up should find the explanation rather than silence.

Why pause went

Docker reports a paused container as running. A paused app stayed listed as healthy: the load balancer kept routing to a server whose processes were frozen and could never answer, monitoring showed a flat line with no explanation, and nothing could release it. appstop does the same job, is visible everywhere as stopped, and can always be started again.

Monitoring is not optional — the node monitors every app it runs for CPU throttling — so stopmonitoring telling a caller "success" meant believing monitoring had stopped while it was still running.

The appstats example was publishing fields we no longer return

It carried the full raw dockerode object. Thirteen of those fields are now dropped before the response reaches the wire:

percpu_usage, usage_in_kernelmode, usage_in_usermode, throttling_data, max_usage, failcnt, pids_stats, num_procs, storage_stats, preread, read, and the container id and name.

The example now matches what the endpoint returns, field for field. Two consequences are called out in the description: memory_stats.stats carries inactive_file only, and percpu_usage is gone — so a consumer using docker's documented "fall back to the length of percpu_usage when online_cpus is 0" trick must read online_cpus directly.

disk_stats is documented properly while here: status is success, partial or error, and a partial reading is a floor rather than a total — charting one as a real value shows a drop that did not happen.

Merge dependency

Depends on RunOnFlux/flux#1777. This spec describes behaviour that only exists once that ships. Merging earlier documents endpoints as removed while they still work.

The customer-facing half is RunOnFlux/flux-docs#16, which rewrites the six game world-upload guides that tell people to use Pause.

Notes for review

• redocly lint passes with one warning: operation-2xx-response on appmonitorstream, which is accurate — that route cannot return a 2xx any more, and documenting one that cannot happen would be worse than the warning.
• docs/index.html is rebuilt, since that is what GitHub Pages serves. Same pinned redocly 2.0.8, 34 insertions, no bundler churn.
• /apps/appmonitor has no entry in this spec at all. It never has. This PR points appmonitorstream callers at it, so it is worth adding — but not inside a deprecation change.