chore: v0.3.3 readme and demo seeder

[mitulgarg]

Summary

• Bump version to 0.3.3 (pyproject.toml, __init__.py).
• Refresh the README's What the Dashboard Shows section to document everything that landed in 0.3.x: groups,
  multi-select / lasso on Topology, group bubbles, the Activity log, inline group editor, and the auto-login token
  injection.
• Add DELETE /api/machines/{id} — idempotent removal of a machine plus its snapshots and queued commands. Useful for
  retiring decommissioned hosts and for cleanup scripts. Returns 204 even if the machine is already gone.
• Add scripts/seed_demo_fleet.py — POSTs synthetic check reports for a 16-machine realistic fleet (H100 / A100 / L40S
  / A10G / RTX 4090 / 3090 / 3080 / 5090 / L4) so the grouping, multi-select, and Activity workflows can be exercised
  without real GPUs. Stable uuid5 ids per hostname so re-runs update rather than duplicate.

Bug fixes

• Seeded machines crashed MachineDetail. DiagnosticCard reads result.issues.length and
  result.recommendations.length directly. The first version of the seeder omitted those fields, so React threw and the
  page rendered blank for any seeded host. Each check is now a complete CheckResult (component, detected, path,
  metadata, issues, recommendations) using the real success / warning / error / not_found status vocabulary.
• Noisy WinError 10054 on every seeded request. urllib's eager socket close conflicted with uvicorn's Windows
  ProactorEventLoop. Added Connection: close to the seeder's POST/DELETE so teardown is clean — no more benign
  ConnectionResetError tracebacks in the dashboard logs.

Files changed

• pyproject.toml, src/env_doctor/__init__.py — version 0.3.2 → 0.3.3
• src/env_doctor/server/routes.py — new DELETE /api/machines/{id} (+ delete SQLAlchemy import)
• README.md — rewritten dashboard features section + auto-login note
• scripts/seed_demo_fleet.py — new (intentionally not referenced from any docs)

Test plan

• pip install -e . and restart env-doctor dashboard.
• python scripts/seed_demo_fleet.py — expect 16 [OK ] HTTP 200 lines and no WinError 10054 traces in the
  dashboard log.
• On the dashboard, click any seeded node (e.g. research-rtx5090-1) — MachineDetail should render with the
  Blackwell sm_120 issue and the pip install nightly cu126 recommendation.
• On Topology, shift+click a few seeded nodes → Assign group → confirm the group bubble appears.
• python scripts/seed_demo_fleet.py --reset — expect each line [OK]; subsequent GET /api/machines should drop
  the seeded entries.
• Re-run --reset immediately — should still succeed (DELETE is idempotent).