fix: lazy flows-DB engine in simulator (v2.0.5)

[damonrand]

Summary

• Stop constructing a SQLAlchemy engine for the flows DB when all
  imbalance data sources are csvTimeseries. The engine was created
  eagerly and unconditionally inside _get_rates_from_config, so a
  placeholder or unparseable env_config["flows"]["dbUrl"] failed at
  URL parse before any simulation logic ran.
• Mirrors the existing rates-DB gating pattern further down the same
  function (rates_config.{live,final}.rates_db is not None).
• Unblocks the standalone CSV-only handoff bundle path, where
  flows.dbUrl is intentionally a "REPLACE_OR_REMOVE…" placeholder.
• When at least one source IS flowsMarketData the engine is still
  constructed; missing dbUrl in that case now raises a clearer
  SystemExit identifying the source of the need.

Bumps 2.0.4 → 2.0.5 (patch). CHANGELOG updated under [2.0.5].

Test plan

• All 14 unit tests pass (PYTHONPATH=src python -m unittest discover --start-directory src)
• Existing integration subtests pass: integrationTestPriceCurve, integrationTestPerfectHindsightLP
• New integration subtest test_integration_csv_only_no_flows_db: drives the existing CSV-only fixture with an unparseable flows.dbUrl ("REPLACE_OR_REMOVE_NOT_USED_BY_SIMULATE") and asserts skypro simulate exits 0. Pre-2.0.5 this would crash at sqlalchemy.create_engine.
• Verified sqlalchemy.create_engine("REPLACE_OR_REMOVE_NOT_USED_BY_SIMULATE") raises ArgumentError on its own — the test exercises a real regression.
• Manual smoke after publish: pip install --upgrade skypro==2.0.5 in a fresh venv, then run a CSV-only bundle without setting flows.dbUrl to a valid URL.

Background

Found while running a standalone bundle exported from skypro-service for an external collaborator. The bundle's README claims the package is self-contained (no DB dependency), but the simulator was crashing on startup because flows.dbUrl had to be a parseable URL even for fully-CSV configs.