spawn autoscale advertises --region and ignores it (ambient region only) #774

Found incidentally while building #772's coverage check, which it made report a
confident false negative.

--region is advertised and ignored

spawn autoscale status --help offers it, inherited from the root:

--region string    Default AWS region (overrides SPORE_REGION/AWS_REGION and the shared config)

It overrides nothing. Demonstrated against an account whose orchestrator is in
us-east-1:

$ AWS_PROFILE=spore-host-infra spawn autoscale status --region us-east-1
⚠️  no autoscale orchestrator found in us-west-2 …

$ AWS_PROFILE=spore-host-infra AWS_REGION=us-east-1 spawn autoscale status --region us-west-2
autoscale orchestrator: spawn-autoscale-orchestrator-staging invoked by … (rate(1 minute))

The flag is ignored in both directions, and AWS_REGION wins even when the flag
contradicts it — the exact inversion of what the help text promises.

Cause

cmd/autoscale_helpers.go builds its config from the ambient chain with no
region applied, in two places:

• getAutoscaler (line ~19) — the EC2/DynamoDB/SQS/CloudWatch clients
• triggerLambda (line ~43) — the immediate-reconciliation invoke

cfg, err := config.LoadDefaultConfig(ctx)   // no config.WithRegion(...)

Other commands do this correctly, so the idiom already exists:

• cmd/cancel.go:160,223 — config.LoadDefaultConfig(ctx, config.WithRegion(region))
• cmd/doctor.go:42 — aws.NewClientWithRegion(ctx, doctorRegion)

Why it matters beyond tidiness

• The DynamoDB scan fails outright in a region with no groups table:
  ResourceNotFoundException: Requested resource not found. That is the visible
  symptom users would hit first.
• triggerLambda invokes spawn-autoscale-orchestrator-<env> by name in the
  ambient region. In a region where the function does not exist, the immediate
  reconciliation silently does not happen — and autoscale launch currently
  treats a trigger failure as a warning, not an error.
• It made Two autoscale orchestrators have polled every minute since February against empty tables #772's new coverage check lie. The probe reported "no autoscale
  orchestrator runs in this account" against an account that has one. I worked
  around it by naming the probed region in the warning so the false negative is
  self-diagnosing, but that is a mitigation, not a fix — a cry-wolf coverage
  warning is the Nothing warns when an account has no TTL-reaper coverage, so the 'everything dies eventually' invariant silently doesn't hold #624 failure, and the workaround only makes it legible.
• An account with per-region deployments (this one has both -production and
  -staging) cannot be addressed deliberately at all.

Fix

Thread the resolved region into both call sites, the way cancel and doctor
already do. Worth checking whether the right move is aws.NewClientWithRegion
for consistency with the rest of the codebase rather than a bare
config.WithRegion.

A test that --region reaches the client would catch the class; today nothing
asserts that any command's --region is honoured, which is why this survived.

Related: #772 (the coverage check this broke), #624 (cry-wolf coverage warnings).