Fixes two related bugs affecting depletion restarts

[GuySten]

Description

Fixes two related bugs affecting depletion restarts that chain continue_timesteps=True across multiple integrate(final_step=False) calls (e.g. running one depletion step at a time, updating model state between steps).

Bug 1: depletion_results.h5 has zero k/reaction rates for every step after the first

Integrator.integrate() writes a placeholder StepResult for the trailing "final simulation" when final_step=False (a zero source rate is passed to the operator, which resets but never runs transport, returning k=0 and zero rates). _get_bos_data() decided whether to reuse prev_res[-1] as beginning-of-step data based only on step_index == 0 — which is true on every restart's Integrator, regardless of whether prev_res[-1] is real data or that zero placeholder. The result: every continuation after the first silently inherited zero reaction rates for its Bateman solve and re-wrote the same zero placeholder back to disk, permanently.

Bug 2: restarted statepoints show a finite k-effective with infinite uncertainty

Separately, write_bos_data() (which triggers a statepoint write) was called unconditionally at the end of integrate(), even when final_step=False. Since no transport solve happened in that case (openmc.lib.reset() was called but openmc.lib.run() never was), the statepoint captured whatever partially-reset internal state was left in the C++ layer — observed as a plausible-looking nonzero keff paired with an infinite standard deviation.

Fix

Added an explicit StepResult.evaluated flag recording whether a step's k/rates came from a real transport/eigenvalue solve or is a final_step=False placeholder. Threaded through save(), distribute(), and the HDF5 read/write path (from_hdf5() defaults to True for pre-existing results files that predate this field, so old files keep working). VERSION_RESULTS bumped to (1, 4).
Integrator._get_bos_data() now checks prev_res[-1].evaluated (not just step_index) before reusing restart data, falling back to a real transport solve when the last saved step was a placeholder.
Integrator.integrate() now only calls write_bos_data() for the trailing step when final_step=True, so no statepoint is written when nothing was actually evaluated.
Testing

Added openmc/deplete/tests/unit_tests/test_deplete_continue_timesteps.py covering:

StepResult.evaluated round-trips through HDF5 save/load, including backward-compatible default for files without the field.
_get_bos_data() correctly distinguishes a genuinely-evaluated restart step from an unevaluated placeholder.
integrate() calls write_bos_data() once (not twice) when final_step=False, and twice when final_step=True.

Also manually verified against the pincell_depletion example using the reporter's exact restart-loop pattern: depletion_results.h5 now has correct, non-zero k/uncertainty at every completed step, and no spurious openmc_simulation_nN.h5 is written for steps that were never evaluated.

Fixes #3973

Checklist

• I have performed a self-review of my own code
• I have run clang-format (version 18) on any C++ source files (if applicable)
• I have followed the style guidelines for Python source files (if applicable)
• I have made corresponding changes to the documentation (if applicable)
• I have added tests that prove my fix is effective or that my feature works (if applicable)