fix: test-mode bypass never writes .completed, so bypassed fits are not resumable

[Jammy2211]

Overview

_fit_bypass_test_mode (the PYAUTO_TEST_MODE=2/3 sampler bypass) writes samples, summary and results but returns without self.paths.completed() — unlike start_resume_fit. A bypass-"completed" search is therefore never seen as complete, so re-running the same script re-bypasses every stage instead of taking result_via_completed_fit. This defeats the resume-representative purpose of PYAUTO_TEST_MODE_SAMPLES (#1378/#1381) and was found by the SLaM resume profiler (autolens_profiling#70), which carries a runtime stopgap.

Plan

• Call self.paths.completed() before the return in _fit_bypass_test_mode (mirrors start_resume_fit; NullPaths/DatabasePaths no-op safely).
• Unit test: after a bypassed fit(), paths.is_complete is true and a second fit() routes through result_via_completed_fit.
• Rider (deferred): remove the stopgap block in autolens_profiling/pipeline_resume/slam_resume.py once the repo's current claim frees — the stopgap is a harmless no-op after this fix.

Detailed implementation plan

Work Classification

Library (PyAutoFit)

Branch Survey

Repository	Current Branch	Dirty?
./PyAutoFit	main	clean

Suggested branch: feature/test-mode-bypass-completed

Implementation Steps

1. autofit/non_linear/search/abstract_search.py::_fit_bypass_test_mode — add self.paths.completed() immediately before return result (after model.unfreeze()).
2. Test in test_autofit/non_linear/ near existing test-mode bypass coverage: run a bypassed fit with DirectoryPaths into tmp, assert paths.is_complete; call fit() again and assert the completed-fit path is taken (e.g. monkeypatch start_resume_fit to raise / spy on result_via_completed_fit).
3. Full test_autofit/ suite.

Key Files

• autofit/non_linear/search/abstract_search.py
• test_autofit/non_linear/ (bypass tests)

Original Prompt

Click to expand starting prompt

Found during slam-resume-profiling (autolens_profiling#70, the test-mode epic's phase 3): in
@PyAutoFit autofit/non_linear/search/abstract_search.py, _fit_bypass_test_mode (the
PYAUTO_TEST_MODE=2/3 path) writes samples, summary and results but returns without calling
self.paths.completed() — unlike start_resume_fit, which marks completion before returning.

Consequence: a bypass-"completed" search is never seen as complete (paths.is_complete false), so
re-running the same script re-bypasses every stage instead of taking result_via_completed_fit.
This defeats the resume-representative purpose of PYAUTO_TEST_MODE_SAMPLES (PyAutoFit#1378/#1381):
the docstring promises "downstream code sees a complete result folder", and .completed is part of
a complete result folder.

Fix is one line — call self.paths.completed() before the return in _fit_bypass_test_mode
(NullPaths/DatabasePaths already no-op safely, matching the surrounding save calls). Add a unit test
asserting paths.is_complete after a bypassed fit, and that a second fit() routes through
result_via_completed_fit.

The autolens_profiling harness (pipeline_resume/slam_resume.py, _install_resume_timers) carries a
runtime patch applying exactly this marker as a stopgap — remove that patch block when this ships.