feat: workout-suggester, weather-aware workouts from a local Gemma

Go + Templ + HTMX app that checks the weather (Open-Meteo, MET Norway
fallback), scores run/ride/walk with plain rules, and has a local
open-weight model (Gemma 4 E2B via any OpenAI-compatible server) write
the plan. Routes are loops from your door (BRouter) plus signposted
OpenStreetMap routes (Overpass), with GPX export.

.env.example

@@ -0,0 +1,18 @@
+# Where your OpenAI-compatible model server lives (oMLX, llama.cpp, Ollama, LM Studio, ...)
+LLM_BASE_URL=http://127.0.0.1:8000/v1
+# Leave empty to use the first model the server lists
+LLM_MODEL=gemma-4-E2B-it-qat-4bit
+# Only if your server requires one
+LLM_API_KEY=
+
+# Prefill the town field
+# DEFAULT_PLACE=Gent
+# ADDR=127.0.0.1:3000
+# LLM_TIMEOUT=3m
+# One or more Overpass endpoints, tried in order
+# OVERPASS_URL=https://overpass-api.de/api/interpreter
+# BRouter plans loops from your door; self-host it or set "-" to turn loops off
+# BROUTER_URL=https://brouter.de/brouter
+# Open-Meteo is open source; point these at your own instance if you like
+# OPEN_METEO_URL=https://api.open-meteo.com/v1/forecast
+# OPEN_METEO_GEOCODE_URL=https://geocoding-api.open-meteo.com/v1/search