Idea: parallelize multi-datasource processing (profile first)

[AlexisJanin]

Datasources are processed sequentially in the wrapper. Each is independent, so there may be wall-clock to gain on patients with many devices.

Worth a look someday, but not obviously worth doing — a few things to check before committing:

• Profile first. Confirm datasource loading is actually the bottleneck (vs. rendering/resampling) on a real, full-size patient.
• Threads vs processes depends on where the time goes (GIL): I/O / pyarrow reads favour threads; Python-heavy formatting would need processes. Note processes are the riskier option here because the app ships as a PyInstaller frozen bundle.
• Touches the sequential progress_callback, output/snapshot ordering, and per-datasource error reporting — so it's not a drop-in.

Park for now; revisit if startup feels slow on real data.

[AlexisJanin]

Decision: idea

Marking as an idea rather than a planned task — genuinely not sure this is worth doing until profiling shows datasource loading is the actual bottleneck on a real patient. Kept for reference, not committed.