Optimizes niceness and cpulimit usage

SCRATCHPAD.md

@@ -14,6 +14,7 @@
 - `FileProperties` now uses one cached `ffprobe -show_format -show_streams -of json` call per source file, and the combined payload was confirmed against the Dragonball asset to satisfy both previous probe call sites fully.
 - Database startup now bootstraps schema only when required tables are actually missing, while version enforcement still runs on ordinary DB-backed context creation.
 - Helper filename and rich-text utilities now use compiled raw regexes plus translate-based filename filtering, with unit coverage for TMDB suffix rewriting and Rich color stripping.
+- Process resource limiting now has explicit disabled/default states in the CLI and requirements, and combined CPU-plus-niceness wrapping now executes as `cpulimit -- nice -n ... <command>` instead of a less explicit prefix chain.
 - FFX logger setup now reuses named handlers, and fallback logger access no longer mutates handlers in ordinary constructors and helpers.
 - The process wrapper now uses `subprocess.run(...)` with centralized command formatting plus stable timeout and missing-command error mapping.
 - Active ORM controllers now use single-query accessors instead of paired `count()` plus `first()` lookups.
@@ -87,15 +88,6 @@
   - Faster contributor onboarding.
   - Easier CI adoption later.
 
-7. Process resource limiting semantics could be clearer
-- [`src/ffx/process.py`](/home/osgw/.local/src/codex/ffx/src/ffx/process.py) prepends `nice` and `cpulimit` directly when values are set.
-- Optimization:
-  - Validate and document effective behavior for combined `nice` + `cpulimit`.
-  - Consider explicit no-limit vs configured-limit states in the CLI and requirements.
-- Expected value:
-  - Fewer surprises in production-like runs.
-  - Easier support for user-reported performance behavior.
-
 ## Open
 
 - Should optimization work focus first on operator-perceived latency, internal maintainability, or correctness-risk cleanup that also has performance upside?