Correct four inaccurate comments in cancel_test.go and rename a constant (closes #33)

Cleanup of the non-blocking findings from the re-review of #6. The tests
were never wrong; only comments that misdescribed the mechanism they
document, and one constant whose name claimed the wrong quantity.

- State the property the walkClock tests rely on (the index load's
  Done cost is bounded and independent of record count) instead of the
  wrong "three consultations" figure.
- Stop the poolUnwind framing from implying every test is bounded at
  two seconds; the three tests that catch their regression only as the
  test binary's timeout now say so.
- Record that hashWorker's results-send abandon branch is reachable
  from the scan path and is covered by TestScanHashWriteFailureUnwindsPool,
  so every cancellation branch has a test.
- Rename walkCancelInFlightDirs to walkCancelInFlightFiles: it is a
  file count (same value).
- Note the deliberate departure from the one-test-file-per-source-file
  convention at the top of the file.

Model: opus-4-8

TODO.md

@@ -29,6 +29,11 @@
 
 # Completed Steps
 
+- correct four inaccurate comments in `cancel_test.go` and rename
+  `walkCancelInFlightDirs` to `walkCancelInFlightFiles` (2026-09-21,
+  branch `issue-33-cancel-test-comments`, closes
+  https://git.eeqj.de/sneak/sfdupes/issues/33)
+
 - fix the lint-image pin comments and `FROM` form in `Dockerfile` and
   `Dockerfile.lint` (2026-08-10, branch `next`, closes
   https://git.eeqj.de/sneak/sfdupes/issues/25): dropped the false