Integrate TRIMP per-sample instead of session-average

[cagnulein]

Summary

• The Bannister TRIMP formula in qfit::save was applied once using the whole-session average HR, instead of being accumulated sample-by-sample.
• Because exp() is convex, a single average-HR pass systematically underweights any workout with HR variability (intervals, surges, sprints) compared to the true per-minute integration — which is how Garmin devices themselves compute training load.
• This is a likely contributor to reports of Garmin Connect's Endurance Score not reflecting real effort for activities uploaded via QZ (see user report: "QZ app - Endurance Score in Garmin Connect").
• Verified against real debug logs/FIT files from 4 test rides (30/30/60/75 min): the new per-sample integration reduces to the same result as before for near-constant-HR sessions, but will score higher for sessions with HR spikes (e.g. one test ride hit 171bpm against a configured max of 162 — a spike the old average-based formula effectively ignored).

Test plan

• Record a steady-effort indoor ride and confirm Training Load/Aerobic TE in the FIT file are unchanged from before (regression check for constant-HR sessions)
• Record an interval/variable-effort ride and confirm Training Load/Aerobic TE now score higher than a steady ride with the same average HR
• Upload both to Garmin Connect and confirm Acute/Chronic Load and Endurance Score look consistent with the actual effort

🤖 Generated with Claude Code

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.