Limit inbound events and absorb shared broadcast bursts

[linkdata]

Incoming events can drive shared-tag broadcasts fast enough to disconnect other users, especially pages with only one element. Pace inbound events per Request and give small pages enough queue space to absorb synchronized bursts.

• Add Jaws.MaxEventRate: 100 records/s by default, zero selects the default, negative disables it. Click, ContextMenu, Input, and JsVar share the budget, including records batched in one frame. Events wait in order; Remove consumes no budget. Use github.com/linkdata/rate.
• Raise the subscription queue floor to 64 while retaining element-count scaling.
• Add the rate argument directly to wire.ReadLoop. Check cancellation after an enabled limiter wait; preserve shutdown and idle-ping behavior.
• Document pacing, bound JavaScript reads, and shared-broadcast bandwidth requirements. The initial event waits 10 ms at the default rate. High-rate browser activity can accumulate delay.

No per-IP aggregate budget is added. Per-Request pacing and burst headroom do not prevent overload when aggregate broadcasts exceed a recipient's sustained capacity; applications should use Dirty/Update coalescing or keep broadcasts within their clients' bandwidth.

Validation: focused transport/request tests, full go test -race ./... and production tests with JAWS_REQUIRE_NODE=1, go generate, gofmt, go vet, staticcheck, golangci-lint, gosec, and go build pass. Reviewed the affected rendered Go docs separately. Tests cover batching, ordering, defaults, disabled/custom rates, idle recovery, cancellation, ping timing, and a continuously reading victim under real WebSocket traffic.

The committed BenchmarkWSBroadcastBurst uses 16 synchronized senders, real WebSockets, and an unthrottled one-element recipient. Server-side pacing is disabled to isolate queue capacity. Linux/arm64; six samples per rate/CPU combination:

go test -run '^$' -bench '^BenchmarkWSBroadcastBurst$' -benchtime=400ms -cpu=1,4 -count=6 -benchmem .
benchstat cap-8.txt cap-64.txt

Events/s per sender	CPUs	Disconnects, cap 8 → 64	Median delivered, cap 8 → 64
25	1	0/6 → 0/6	100% → 100%
25	4	4/6 → 0/6	70.05% → 100%
400	1	2/6 → 0/6	100% → 100%
400	4	6/6 → 0/6	10.83% → 100%

Total recipient disconnects fell from 12/24 to 0/24. Benchstat reports the 400/s, 4-CPU delivery improvement at p=0.002. An exploratory 16/32/64-capacity sweep also passed at 32; 64 provides additional burst headroom. These short runs measure burst tolerance, not a guarantee against sustained overload. Time/op includes client pacing; allocations include more completed delivery work when the recipient stays connected.

Fixes #536.