Update android_handler.py to support rule-based objection handling with Supabase templates.

Requirements:
1) Imports
   - from supabase import create_client
   - import os, re, datetime as dt
   - from zoneinfo import ZoneInfo
   - from tools import (
       resolve_agreement_id, log_message, open_thread, close_thread,
       schedule_followup, book_appt_request, get_equity_snapshot
     )
   - Handle OPENAI_API_KEY optionally: if missing, skip OpenAI and use template-only replies.

2) New helpers
   a) get_client():
      - Create a single Supabase client from SUPABASE_URL and SUPABASE_SERVICE_KEY.
   b) fetch_rules_and_templates(client) -> (rules, templates):
      - rules: select * from objection_rule order by priority asc;
      - templates: select slug, body from reply_template where is_active = true;
      - Return two dicts: rules = [{label, match_any, template_slug, priority}], templates = {slug: body}.
   c) fetch_customer_context(agreement_number) -> dict:
      - Query v_agreement_latest to get: first_name, last_name, registration, make, model, agreement_number.
      - Return a dict with keys: first_name, last_name, vehicle_model (combine make + model), registration, agreement_number.
   d) personalise(template_str, ctx) -> str:
      - Replace {{first_name}}, {{last_name}}, {{vehicle_model}}, {{registration}} safely (fallback to empty if missing).
   e) match_rule(text, rules) -> template_slug | None:
      - Lowercase text; for each rule in priority order, if ANY keyword in rule.match_any is a substring, return template_slug.
      - Use simple normalised matching (lowercase; strip punctuation).
   f) ok_to_reply():
      - Use Europe/London time; return True if 06:00–23:59:59.

3) Reply logic (in the message processing loop)
   - After detecting a new inbound message:
     - ctx = fetch_customer_context(agreement_number)
     - rules, templates = fetch_rules_and_templates(client)
     - slug = match_rule(inbound_text, rules)
     - if slug and slug in templates:
         reply_text = personalise(templates[slug], ctx)
       else:
         # Fallback to OpenAI (if key present) using Lia’s UK system prompt
         # System prompt must emphasise UK English dealership tone and “no pricing promises”.
         # If OpenAI not available, reply_text = personalise(templates.get("ack_hello", "Hi {{first_name}}, it’s Lia at Lincoln Audi. How can I help?"), ctx)
     - Respect ok_to_reply(); if False, schedule_followup for next day 06:05 UK and do not send now (just log deferral).
     - Otherwise:
         - log_message(agreement_number, channel, "outbound", reply_text)

4) Lia’s system prompt (use for OpenAI path)
   - System: "You are Lia Carter, Customer Loyalty Advisor at Lincoln Audi. Always write in natural UK English with warm, conversational phrasing typical of a British dealership professional. Be helpful, relaxed and human. Never promise pricing or 'same payment'. Invite a short visit or a quick call if needed. Keep replies short."
   - User content: include the inbound_text plus a compact summary of ctx (first_name, vehicle_model, registration).

5) Minimal safety:
   - If template 'book_service' was used, DO NOT fake-book. Ask for rough mileage + preferred dates + courtesy-car. (We already have that text.)
   - If price-related keywords matched ('too_expensive'), DO NOT quote; invite short visit/review and offer to check options.

6) Console logs:
   - Print: matched rule label (or "no rule → OpenAI"), chosen template slug, and that outbound was logged.

Deliverable:
- Modify android_handler.py accordingly.
- Do not remove existing polling/loop; just integrate this logic where replies are generated.
- When done, print: UPDATED android_handler.py (rules + templates integrated)