[orian]

Summary

• load web manifest compositions concurrently instead of composing each role/environment serially
• load per-node dump models through a bounded pool of up to 32 workers, allowing a 30-node fleet to start all loads together
• preserve deterministic manifest and filename result order regardless of completion order
• print startup progress to stderr at bounded intervals, including worker count, completed tasks, final success, and failure count
• keep single-config and drift command behavior unchanged
• document parallel startup and progress behavior

Example startup output:

web: loading 30 dump models in parallel (30 workers)
web: loading dump models: 1/30 complete
web: loading dump models: 3/30 complete
...
web: loaded 30 dump models

End-to-end coverage

• start hclexp web in a subprocess against 30 real node dumps
• verify parallel progress is emitted on stderr, stdout remains empty, and the browser reaches its serving state
• build a 30-node dump browser and verify every node is mounted
• use a worker barrier to prove all 30 jobs start concurrently before any finishes
• verify results and errors retain deterministic input order

Verification

• go test ./... (1365 tests)
• go test -race ./internal/... ./cmd/... (1351 tests)
• go vet ./...
• go build ./...
• gofmt -s -l .
• git diff --check