GEOPY-2910: Reduce chunking of sensitivities for TEM inversions

[domfournier]

GEOPY-2910 - Reduce chunking of sensitivities for TEM inversions

[Copilot]

Pull request overview

This PR updates the Dask-based TDEM sensitivity assembly to reduce sensitivity “chunking” by stacking field-derivative solves across each parallel block, aiming to lower overhead and improve inversion performance/memory behavior.

Changes:

• Disables get_parallel_blocks(..., optimize=True) re-splitting to avoid extra block/chunk fragmentation.
• Refactors get_field_deriv_block to build a single dense ATinv_df_duT_v array per block (instead of per-chunk lists) and solve in a stacked call.
• Updates compute_rows to slice the new per-block derivative array using a running column offset (colm_count).

Comments suppressed due to low confidence (1)

simpeg/dask/electromagnetics/time_domain/simulation.py:482

• When local_ind is empty, this branch continues without advancing colm_count. Since later iterations slice field_derivs[block_ind] using colm_count, subsequent chunks will be misaligned (wrong columns used for derivatives), producing incorrect sensitivity rows.

 if len(local_ind) < 1:
row_block = np.zeros(
(len(ind_array[1]), simulation.model.size), dtype=np.float32
)
rows.append(row_block)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.