test: loosen wall-clock tolerances in download handler tests (CI flake)

[AronPerez]

Summary

• loosen only the tight wall-clock upper bounds in the download handler tests
• keep timeout inputs, lower bounds, cleanup checks, and logic assertions unchanged
• document that the ceilings allow for scheduling jitter on shared CI runners

Context

This addresses two flakes on PRs unrelated to the test file:

• run 29945960477: test_handle_action_timeout_bounds_browser_download_handler_drain took about 0.214s against a 0.2s ceiling
• run 29956252756: test_handle_action_download_in_flight_request_does_not_extend_custom_timeout took about 0.181s against a 0.1s ceiling

These tests validate bounded timeout behavior and whether an in-flight request extends the configured timeout; exact sub-second wall-clock scheduling is not part of that contract. The looser upper ceilings preserve those properties while tolerating shared-runner jitter.

Validation

• ran tests/unit/test_download_file_action_handler.py five times: 77 passed on every run
• uv run pre-commit run --all-files

[AronPerez]

Third occurrence of this flake family in ~7h, again on a PR that doesn't touch the file: run 29970786304 (PR #7679, a wording-only change) failed test_handle_action_download_without_explicit_timeout_has_bounded_in_flight_extension with assert 0.2916 < 0.15. That's three distinct tests in this file tripping wall-clock bounds on CI runners (0.2s, 0.1s, 0.15s). Each occurrence costs a sync-queue merge slot + a ~13min retrigger cycle.

[AronPerez]

Fourth occurrence: run 29972647695 (PR #7679's retry) failed test_handle_action_download_in_flight_request_does_not_extend_custom_timeout again — assert 0.184 < 0.1. The family is now failing most runs, and the sync queue's oldest PR is held on it. This fix unblocks the pipeline.