fix(server): honor HYPERGLASS_WORKERS for web worker count

[WilhelmZA]

Problem

HyperglassSettings.workers is a computed property hardcoded to cpu_count(2)
(2x logical cores) whenever debug is off, and hyperglass/main.py duplicates
that same inference inline. There is no settings field backing it, so the
HYPERGLASS_WORKERS environment variable is silently ignored.

On a 24-core host this starts 48 uvicorn workers regardless of intent — each
a full process holding the app plus its own Redis connections — with no
supported way to cap it. For a looking glass that is far more than needed.

Fix

• Add a real workers settings field (HYPERGLASS_WORKERS), Optional[int],
  default None.
• Replace the unused workers property with worker_count, which honors the
  field (clamped to >= 1) and otherwise keeps the previous behavior: 1 in
  debug mode, else 2x CPU cores.
• run() now uses Settings.worker_count instead of duplicating the inference.

HYPERGLASS_WORKERS	debug	workers
unset	false	2x cores (unchanged default)
unset	true	1 (unchanged)
4	either	4
0	either	1 (clamped)

Notes

• Self-contained: touches only models/system.py and main.py, independent of
  any other in-flight work, and applies cleanly to the current main.
• Backward compatible — the default is unchanged when HYPERGLASS_WORKERS is
  unset.
• Adds unit tests for worker_count.