chore: implement a project overhead scale test

[savme]

Summary

Phase one of #596.

• Adds a scale build-tag test in test/scale/ that measures per-project memory, goroutine, and etcd watcher overhead against a live Milo API server
• Captures before/after snapshots (heap, sys, goroutines, etcd watchers) with pprof profile export to dev/profiles/
• Adds task test:scale and excludes generated profiles from git
• Uses existing dev cluster setup and go test

This establishes a baseline (the test does not define any thresholds) to help us quantify the current per-project costs before implementing any architectural changes.

How to run

task dev:setup                     # provision dev infra 
task test:scale                    # 20 projects (default)

# or to run with n projects
SCALE_PROJECTS=50 task test:scale

Profiles are written to dev/profiles/heap-{N}p-{before,after}.pb.gz and can be compared with go tool pprof.

Test plan

• task test:scale completes against a task dev:setup environment
• go test ./... (without -tags scale) is unaffected

Example Run

Log

=== RUN   TestProjectControlPlaneOverhead
    project_overhead_test.go:107: connected to milo apiserver at https://localhost:30443
    project_overhead_test.go:108: port-forward milo-system/svc/etcd: localhost:63269 -> 2379
    project_overhead_test.go:112: resources: 52 listable types
    project_overhead_test.go:114: [before] Δgoroutines=1661 apiserver[heap=125.1MiB sys=183.7MiB goroutines=1661] etcd[watchers=54]
    project_overhead_test.go:114: [before] Δgoroutines=1 apiserver[heap=129.9MiB sys=183.7MiB goroutines=1662] etcd[watchers=54]
    project_overhead_test.go:114: [before] Δgoroutines=9 apiserver[heap=134.8MiB sys=183.7MiB goroutines=1653] etcd[watchers=54]
    project_overhead_test.go:114: [before] Δgoroutines=1 apiserver[heap=111.4MiB sys=183.7MiB goroutines=1654] etcd[watchers=54]
    project_overhead_test.go:115: latency before: n=52 errors=0 p50=1ms p90=2ms p99=3ms max=8ms
    project_overhead_test.go:160: bootstrap: 104/1040
    project_overhead_test.go:160: bootstrap: 208/1040
    project_overhead_test.go:160: bootstrap: 312/1040
    project_overhead_test.go:160: bootstrap: 416/1040
    project_overhead_test.go:160: bootstrap: 520/1040
    project_overhead_test.go:160: bootstrap: 624/1040
    project_overhead_test.go:160: bootstrap: 728/1040
    project_overhead_test.go:160: bootstrap: 832/1040
    project_overhead_test.go:160: bootstrap: 936/1040
    project_overhead_test.go:160: bootstrap: 1040/1040
    project_overhead_test.go:119: [after] Δgoroutines=21782 apiserver[heap=333.1MiB sys=484.9MiB goroutines=21782] etcd[watchers=1054]
    project_overhead_test.go:119: [after] Δgoroutines=1097 apiserver[heap=341.0MiB sys=484.9MiB goroutines=20685] etcd[watchers=1054]
    project_overhead_test.go:119: [after] Δgoroutines=0 apiserver[heap=348.8MiB sys=484.9MiB goroutines=20685] etcd[watchers=1054]
    project_overhead_test.go:119: [after] Δgoroutines=8 apiserver[heap=361.6MiB sys=484.9MiB goroutines=20693] etcd[watchers=1054]
    project_overhead_test.go:119: [after] Δgoroutines=5 apiserver[heap=371.1MiB sys=484.9MiB goroutines=20698] etcd[watchers=1054]
    project_overhead_test.go:120: latency after:  n=52 errors=0 p50=1ms p90=1ms p99=3ms max=5ms
    project_overhead_test.go:123: results: projects=20 resources_per_project=52
    project_overhead_test.go:124:   before:      apiserver[heap=111.4MiB sys=183.7MiB goroutines=1654] etcd[watchers=54]
    project_overhead_test.go:125:   after:       apiserver[heap=371.1MiB sys=484.9MiB goroutines=20698] etcd[watchers=1054]
    project_overhead_test.go:126:   per-project: apiserver[heap=13.0MiB sys=15.1MiB goroutines=952] etcd[watchers=50]
    project_overhead_test.go:128: profiles: saved milo/dev/profiles/heap-20p.pb.gz
    project_overhead_test.go:128: profiles: saved milo/dev/profiles/goroutine-20p.txt
--- PASS: TestProjectControlPlaneOverhead (73.27s)