Finish landing ioxide.timer: pack it, and give it a Playground sample

[MDA2AV]

ioxide.timer landed in #213 but did not land completely: CI never packed it, and it has no Playground sample. Both here.

ci: pack ioxide.timer

The pack steps in the release workflow are one per project and hand-written, so adding a project to ioxide.slnx builds it but does not ship it. That is what happened to 0.7.211 — the SubmitTimeout in core went out, the client over it did not. Checked the rest against the tree while there; every project carrying a PackageId now has a pack step.

No version change is needed: ioxide.timer 0.7.211 was never published, so the next release picks it up at the version its siblings already carry.

Playground/Clients/Timer

Every other package has a sample that is a complete server, a row in bench/samples.tsv, and a pane on the site generated from that same file. A package without one is a package a reader has to learn from its doc comments.

GET /<ms> answers after that many milliseconds:

dotnet run -c Release --project Playground/Clients/Timer
curl http://127.0.0.1:8080/50     # -> 50ms, after 50ms

It holds one RingTimer for the connection and re-arms it per request. That is the shape a caller is meant to copy, and it carries the one constraint worth knowing: a timer holds a single RingOpSource, so it carries one wait at a time — something waiting on several deadlines at once wants a timer each.

It also shows the thing a caller would otherwise get wrong. An expired timeout completes with -ETIME, which is io_uring reporting success, so the check is RingTimer.Expired(result) and not result >= 0. The sample writes errno N into the body when it is anything else, so a mistake there would be visible rather than silent.

The useTaskDelay knob runs the same server with Task.Delay in place of the ring — same response, only the wait changes — so the off-reactor hop it costs is measurable rather than asserted, the way Tcp/Hop is runnable next to Tcp/Raw. The delay comes from the request, so it is clamped; an unbounded one would let a client hold a connection all day.

Registered as h1 on :8080 with PLAYGROUND_DELAY_MS=0, which prices the wait itself against Tcp/Raw rather than measuring a sleep. bench/any.sh --list reports it runnable.

Also here

README.md's client package table was missing ioxide.timer entirely. Added.

The site's example panes are regenerated from the samples, so scripts/gen-docs-panes.py gains the timer entry and docs/index.html gains the tab, the pane and its three CSS rules. Only the new pane changed — the rest were already in sync, and both generators report "already up to date" on a second run.

Verified

Against the running sample: / waits the default 25ms, /50 waits 50ms, /200 waits 200ms, a non-numeric path falls back, the ceiling holds at 60s, and three keep-alive requests on one connection each wait their own 40ms rather than sharing a deadline. The PLAYGROUND_TASK_DELAY=1 arm serves identically. Full solution builds clean, 0 warnings.