Add parallel FYRE OpenShift benchmarks with 40 GB disks

[lucarlig]

The standalone FYRE API fixes each Ubuntu VM at a 250 GB root disk, while FYRE OpenShift defaults every worker to a 250 GB root disk plus two 200 GB data disks. This adds OpenShift-backed benchmark profiles that set every master and worker root disk to 40 GB and explicitly request no worker data disks.

The quota-sized openshift-2v2-parallel.yaml profile runs all eight one-hour measurements concurrently: built-in and external dataplanes at 125, 250, 500, and 1,000 users. Every measurement has an independent 2 vCPU/2 GiB target limit, 1.5 vCPU/1.25 GiB Locust pod with one master and three workers, and 1.5 vCPU/1.375 GiB zero-delay Fast Time pod. The target pods reserve 1.75 vCPU/1.75 GiB each so all eight fit with scheduling headroom on a 16 vCPU/16 GiB target worker. Locust and Fast Time use separate 14 vCPU/12 GiB workers. The helper-pressure gate rejects results if load generation or Fast Time becomes the bottleneck.

Both lanes use the same modern 2026-07-28 Locust client, server/discover startup, and repeated balanced tools/call traffic across the same six Fast Time tools. Each request carries identical per-request client metadata. Only the public MCP route and target implementation differ.

The CLI owns the complete OpenShift lifecycle through the FYRE API, stores state under CF_INTEGRATION_DIR, downloads artifacts before cleanup, redacts credentials, and supports run, status, and destroy with short aliases. Re-running load fyre run with the same run ID resumes an interrupted OpenShift campaign from its saved manifest. Reports include measured-window gateway memory, helper-pressure validation, CSV/JSON, a Slack-ready PNG, and a self-contained Markdown report with the request mix and Mermaid architecture.

The fully parallel three-worker profile requests 740 GB total disk: one fixed 500 GB FYRE API VM, three 40 GB masters, and three 40 GB workers. A live SVL probe confirmed that base_disk_size: "40" plus additional_disk: [] is accepted and produces 40 GB worker roots with no data disks. A full live campaign is provisioning as cf-ocp-2v2-parallel-rtp1 (cluster 2-289711); its supervisor is independently managed and will resume from the same run ID if interrupted.

Validation:

• cargo test --locked --all-targets (435 passed)
• cargo clippy --locked --all-targets --all-features -- -D warnings
• PYTHONPATH=benchmarks/fyre python3 -m unittest benchmarks/fyre/test_campaign.py benchmarks/fyre/test_openshift_campaign.py (31 passed)
• Live FYRE API validation of 40 GB master/worker roots and empty worker data disks
• Live full campaign accepted with seven cluster VMs and all eight benchmark measurements configured in parallel