[T-Crypt]

A quality pass over the repo. Two README claims contradicted the code, one
collector test failed on the GPU box, and the /api/requests endpoint had no
unit test.

Docs

• The HTTP API table documented /api/requests?range=…. The handler reads
  since (epoch seconds) and limit (default 1000, clamped to 1000), so a
  range=… request is ignored and returns unfiltered rows. The table row now
  reads ?since=…&limit=….
• The README said "the two test_hostinfo.py skips". The Windows ctypes
  backend holds three tests, so a Linux run skips three. The sentence now
  describes the gates with no count.

Tests

• collector/test_hostinfo.py: test_processes_shape failed on the GPU box
  when a tiny launcher process matched an engine name. rss_gb is rounded to
  2 dp in GB, so a process of about 2 MB reports 0.0, which failed the > 0
  assertion. Both Linux process tests now accept >= 0.
• collector/test_speculum.py: three tests for the /api/requests contract.
  They pin the default (since 0, limit 1000), the explicit pass-through,
  and the 400 for a non-integer value or limit < 1.

Test output

python3 -m py_compile collector/*.py                         # OK
python3 -m unittest discover -s collector -p "test_*.py"     # Ran 109 tests, OK (skipped=3)
node smoke/ui-primitives.mjs                                 # ALL PASS
node smoke/shell.mjs                                         # ALL PASS

The repo and CI configure no linter. The documented parse check is
py_compile. flake8 --select=E9,F63,F7,F82 collector/ is clean.

Issues

gh issue list --state all is empty. The repo has no open or closed issues,
so there was nothing to triage.

No dependency changes and no drive-by refactors.