refactor: align transform layer with template methodology

Three deviations from the quart_saas_boilerplate methodology corrected:

1. Fix dim_cities LIKE join (data quality bug)
   - Old: FROM eurostat_cities LEFT JOIN venue_counts LIKE '%country_code%'
     → cartesian product (2.6M rows vs ~5500 expected)
   - New: FROM venue_cities (dim_venues) as primary table, Eurostat for
     enrichment only. grain (country_code, city_slug).
   - Also fixes REGEXP_REPLACE to LOWER() before regex so uppercase city
     names aren't stripped to '-'

2. Rename fct_venue_capacity → dim_venue_capacity
   - Static venue attributes with no time key are a dimension, not a fact
   - No SQL logic changes; update fct_daily_availability reference

3. Add fct_availability_slot at event grain
   - New: grain (snapshot_date, tenant_id, resource_id, slot_start_time)
   - Recheck dedup logic moves here from fct_daily_availability
   - fct_daily_availability now reads fct_availability_slot (cleaner DAG)

Downstream fixes:
- city_market_profile, planner_defaults grain → (country_code, city_slug)
- pseo_city_costs_de, pseo_city_pricing add city_key composite natural key
  (country_slug || '-' || city_slug) to avoid URL collisions across countries
- planner_defaults join in pseo_city_costs_de uses both country_code + city_slug
- Templates updated: natural_key city_slug → city_key

Added transform/sqlmesh_padelnomics/CLAUDE.md documenting data modeling rules,
conformed dimension map, and source integration architecture.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

transform/sqlmesh_padelnomics/models/serving/planner_defaults.sql

@@ -13,7 +13,7 @@ MODEL (
   name serving.planner_defaults,
   kind FULL,
   cron '@daily',
-  grain city_slug
+  grain (country_code, city_slug)
 );
 
 WITH -- Real city-level benchmarks from Playtomic