Consolidate DB access behind DatabaseInterface and add orchestrator

[thedeepaksengar]

Consolidated leaked DB queries. There were 33 raw pymongo calls scattered
across 6 files (functions.py, custom_decorators.py, info.py, queue_manager.py,
ProVe_main_service.py, background_processing.py, ProVe_heuristic_service.py).
All of them now go through methods on MongoDBHandler. Added 16 new methods
(get_latest_status_by_qid, get_html_by_task_id, upsert_summary_by_id,
log_usage, etc). Also folded StatsDBHandler and TMPStatsDBHandler subclasses
back into MongoDBHandler since they were just the same class pointing at
different DB names — the new handler lazily opens the usage DBs on demand.

Added the DB orchestrator / interface layer. New prove_shared/database/
subpackage with:

• interface.py — DatabaseInterface that MongoDBHandler (and the future
  PostgreSQLHandler) implement
• postgres.py — stub for the Postgres implementation, every method raises
  NotImplementedError with notes on what the SQL equivalent should look like
• orchestrator.py — DatabaseOrchestrator + get_database() factory, reads
  the active backend from config.yaml

All 6 callers now do db = get_database() instead of MongoDBHandler().
Switching DBs is config-only:

database:
  primary: mongo         # or postgres
  fallback: none         # or mongo / postgres
  mode: single           # or dual-write (during migration)
  auto_fallback_on_read: false

Review-feedback changes on top of the initial PR.

• Moved mongo_handler.py into database/ alongside postgres.py for symmetry. Updated every import site; nothing references the old path.
• Restored full Google-style docstrings (Args/Returns/Raises) on every method, new and existing — so Sphinx/pdoc autodoc picks them up.
• Renamed the module-level mongo_handler variable to database_handler everywhere (it may now hold a PostgreSQLHandler or an orchestrator, so the old name was misleading). Same for local mongo_status /mongo_statuses → status_doc / status_docs.

Added unit tests. 116 tests, all passing:

• test_mongo.py — Mongo-specific unit tests for all 16 new methods (mock pymongo, verify call translation) plus the refactored requestItemProcessing helper.
• test_database_contract.py — backend-agnostic contract tests via a pytest fixture. Currently parametrised over Mongo only; when Postgres lands we add its fixture and the whole suite auto-runs against both.
• test_orchestrator.py — single / dual-write / read-fallback routing, log-and-continue semantics, claim-op primary-only, factory caching.
• test_postgres.py — stub interface compliance + every stub method raises NotImplementedError with a message pointing at itself.

Bugs fixed along the way.

• Race condition in get_summary — the insert-or-update branch now uses upsert=True atomically.
• Race condition in retry_processing — read-modify-write of retry_count replaced with $inc.
• process_top_viewed_items / process_pagepile_list were passing 'top_viewed' / 'pagepile_weekly_update' as the queue arg instead of request_type
• custom_decorators.py was opening a fresh Mongo connection on every HTTP request.
• Latent type-hint bug found by the new tests: pymongo.collection is a module, not a class (the class is pymongo.collection.Collection). Fixed throughout mongo.py.

[NathanGavenski]

Hey @thedeepaksengar great PR, let me know when you go through the comments so we can have a small meeting