[paultristanwagner]

All code in this PR was written by an AI agent (Claude Opus 5.5) working with me; I directed it and tested the results on my own account. Feel free to take, change or drop any part of it.

Part of #73.

Adds search_activities and add_exercise tools to log exercise.

The app logs exercise with POST /api/v2/add_exercise:

{"exercise": {"exerciseId": null, "type": "Exercise", "activityId": 1342,
              "activitySpecId": 370, "minutes": 15.0, "weight": 176.4,
              "calories": -42.0, "calorieOverride": false, "day": …,
              "order": …, …},
 "auth": {…}, "config": {"call_version": 1}}

The app works out calories itself, from activity intensity data that
isn't in the activity lists. This change has the caller pass the calorie
burn and sets calorieOverride, like entering calories by hand in the app.
Verified against the live API on 1900-01-02:

• activitySpecId can be left out (it's stored as 0)
• weight (body weight in lb) is required: without it the server
  answers 500. It's taken from the profile's weightInKG pref.
• activity ids from both catalogs the app's picker merges work:
  get_activity_forest (about 200 activities, ids 1135 and up) and
  all_legacy_activities (about 1100 more specific ones, ids 1-1134,
  e.g. "Walking 3.0 mph, moderate"). The id ranges don't overlap.

Exercises appear in get_diary as "type": "Exercise" entries. Deleting
them needs delete_entries to match exerciseId (#74).

Changes:

• client: get_activities() merges both catalogs.
  add_exercise(activity_id, minutes, calories, day, diary_group) sends
  the burn as a negative number with calorieOverride: true, plus the
  profile weight.
• server: search_activities (every query word must appear in the
  name) and add_exercise. Its description asks the model to estimate the
  burn from the MET value when the user doesn't know it.
• README: tool table rows.

https://claude.ai/code/session_01MCJFDo71qoSqTRtcGfkpXJ