perf: Skip VRAM, I/O, model.info in test mode 2+ and add PYAUTO_DISABLE_JAX

[Jammy2211]

Summary

When PYAUTOFIT_TEST_MODE >= 2, skip expensive operations that aren't needed for smoke testing: VRAM estimation (16s), pre/post-fit file I/O (10s), model.info formatting (7s), and result.info formatting (5s). Also adds PYAUTO_DISABLE_JAX=1 env var to force use_jax=False in the Analysis base class.

Combined with the earlier PYAUTO_WORKSPACE_SMALL_DATASETS and PYAUTO_DISABLE_CRITICAL_CAUSTICS env vars, this reduces smoke test runtime from ~100s to ~22s per modeling script.

API Changes

Added one environment variable: PYAUTO_DISABLE_JAX=1. When set, Analysis.__init__ forces use_jax=False regardless of what the caller passes.

In test mode 2+, model.info returns a stub string and result.info returns a stub string. print_vram_use() is a no-op. search.fit() skips pre-fit output, post-fit output, and all file I/O in the bypass path. These are internal behavioural changes that only apply when the test mode env var is explicitly set.

See full details below.

Test Plan

• PyAutoFit unit tests pass (1198 passed)
• imaging/modeling.py completes in ~22s with all env vars
• interferometer/modeling.py completes in ~22s with all env vars
• Without env vars, original behaviour preserved

Full API Changes (for automation & release notes)

Added

• Environment variable PYAUTO_DISABLE_JAX=1 — forces use_jax=False in Analysis.__init__

Changed Behaviour (test mode 2+ only)

• Analysis.print_vram_use() — returns immediately (no-op)
• AbstractSearch.fit() — skips pre_fit_output(), post_fit_output(), modify_after_fit()
• AbstractSearch._fit_bypass_test_mode() — skips save_samples(), save_results(), save_results_combined(), paths.completed()
• AbstractPriorModel.info — returns stub string
• text_util.result_info_from() — returns stub string

Migration

None — all changes are gated behind PYAUTOFIT_TEST_MODE >= 2 or PYAUTO_DISABLE_JAX=1.

🤖 Generated with Claude Code