Fix multi-sport activity import (#289)

[philippeauriach]

Multi-sport activities (triathlon / brick sessions) are stored by Garmin as a single FIT file containing multiple session messages plus child sessions for transitions. Previously the importer wrote every session under the bare parent activity_id, so the last session silently overwrote all the others and laps/records/HR zones/devices ended up attached to the wrong activity (or dropped entirely when the hr_zones_timer processing order created empty rows that blocked the real lap writes).

This change detects multi-sport files by checking len(fit_file.session) > 1 and, when detected:

• creates one activity row per session with a suffixed id <parent>_<n> (e.g. 22638574127_1 ... _5 for swim→T1→run→T2→swim)
• derives each session's stop_time from start_time + total_elapsed_time (Garmin reuses the parent start_time as every child's timestamp, which would otherwise result in stop_time <= start_time)
• partitions laps by session using each session's num_laps and rewrites lap_num to be session-local (0..N per session), so every child has its own contiguous lap sequence
• partitions records (GPS trackpoints) by matching their timestamp against each session's time window
• remaps session/lap HR-zone writes to the correct child activity_id
• associates device entries with every child activity

_write_lap_entry was also switched to ActivityLaps.s_insert_or_update so lap data correctly coalesces with the hr_zones_timer rows that the generic message dispatcher creates earlier in the import; the same stop_time correction applies when the file is multi-sport.

Single-session FIT files go through the exact same code path as before.

Extends and supersedes the single-session subset fixed in #310 (same root cause for stop_time, lap-ordering race, and the add-vs-insert-or-update issue).

Fixes #289

[tcgoetz]

Please re-target this Pr against the develop branch.

[philippeauriach]

hello @tcgoetz , base branch update to develop! thanks