delete parallel calculation folder and add a summarized markdown in doc folder

doc/parallel_ansys_sweeps.md

@@ -0,0 +1,42 @@
+# Running a parameter grid across parallel Ansys sessions
+
+This documents the pattern originally in `machine_design/parallel_calculation/`
+(`runner_parallel.py` , `calculate_combination.py`, added by Jan Laksar), kept
+here after that folder was removed in issue #21.
+
+## Problem
+
+A single Ansys Electronics Desktop session can only run one design at a time,
+so sweeping a grid of operating points (for example Id/Iq current setpoints)
+sequentially in one session is slow. Running several independent Desktop
+sessions in parallel, each solving its own share of the grid, cuts wall time
+roughly by the number of workers.
+
+## Method
+
+1. Split the full grid of tasks into `N_WORKERS` contiguous chunks.
+2. For each worker, copy the base `.aedt` project file to its own file before 
+   opening it. Ansys locks a project file while it is open, so every worker 
+   needs its own copy rather than sharing one file.
+3. Start `multiprocessing.Pool(N_WORKERS)` and run one worker function per
+   chunk. Each worker opens its own `Desktop(non_graphical=True,
+   new_desktop=True, close_on_exit=False)` and `Maxwell2d` instance against
+   its private project copy, so the workers do not interfere with each
+   other's Ansys session.
+4. Inside a worker, solve each task with `analyze_setup(setup_name,
+   use_auto_settings=False, cores=1, tasks=1)`, pinning each solve to a single
+   core. Several workers are already running concurrently, so letting each one
+   grab multiple cores would oversubscribe the machine.
+5. Extract the quantities of interest with
+   `post.get_solution_data(expressions=..., primary_sweep_variable="Time")`,
+   then average over `data_real(expr)[:-1]` (dropping the last point, which
+   repeats the first point of the periodic waveform).
+6. The main process collects every worker's results and writes them out
+   once, as both a plain CSV and an HDF5 file.
+
+## Example
+
+The removed `runner_parallel.py` swept a 16x16 grid of `Id`/`Iq` values (0.0
+to 3.2 A in 0.2 A steps) and recorded flux linkages, inductances, and torque
+for each point, with `calculate_combination.py` doing the per-point solve
+and extraction described above.