feat: multi-and-per-flow rates schema (v2.2.0)

[damonrand]

Summary

Three additive schema features for simulate.yaml. All backward-compatible — existing configs unchanged. Closes the structural Axle-premium leak on two-MPAN BSCP550 sites and removes cut-and-paste duplication for multi-settlement workflows.

• enabled: false on a simulation drops it pre-schema, so externally-managed scenarios (e.g. Monte-Carlo runs whose outputs are produced outside skypro) can sit alongside live ones in the same yaml. Removes the rebuilder's temp-yaml staging workaround.
• Per-flow imbalanceDataSourceOverride inside RatesFiles flows. Each flow can override the rates-block-level imbalanceDataSource via a new dict shape; legacy list shape stays valid. Closes the HMCE Apr-26 BSCP550 ~£500–700/mo over-attribution where site-MPAN flows absorbed Axle premium intended only for BESS-MPAN flows.
• Multi-finals declares N settlement variants per simulation. Mutually exclusive with the legacy final: field (peak/peaks precedent). Fanned out at parse time into one expanded sim per variant; CSV paths get an automatic variant suffix when not using $_SIM_NAME.

Commits (preserve as merge commit, not squash, so CHANGELOG SHAs resolve)

• 7b7caa6 docs: design briefs for the two non-trivial changes
• ea2d492 feat: accept enabled: false on simulations
• ca77995 feat: per-flow imbalanceDataSource override on RatesFiles
• 2188e35 feat: multi-final fan-out
• c54c7cc chore: bump version to 2.2.0
• 3502780 docs: CHANGELOG entry + CLAUDE.md schema reference + mark proposals IMPLEMENTED

Compatibility

• Legacy single-final configs unchanged. Verified by integrationTestPriceCurve, integrationTestPriceCurveMultiPeak, integrationTestPerfectHindsightLP — bit-identical LP output within tolerance 0.01.
• Legacy list-shape rate files continue to parse and behave identically.
• ratesDB source rejects per-flow overrides with a clear error (override only supported with the YAML files source).

YAML reference

# 1. enabled:false annotation
simulations:
  hmce.202512.imb.mc-reopt:
    enabled: false                # dropped pre-schema, outputs preserved as-is

# 2. per-flow imbalance override (apply symmetrically to live + final)
files:
  gridToBatt: [ a.json, b.json ]                # legacy — inherits block-level
  solarToGrid:                                  # dict shape with override
    rates: [ a.json, b.json ]
    imbalanceDataSourceOverride: *imbSrc_plain

# 3. multi-finals fan-out (mutually exclusive with `final:`)
rates:
  live:  *ratesLive_basecase
  finals:
    fullfcl:      *ratesFinal_fullfcl
    trio_imbflex: *ratesFinal_trio_imbflex

Test plan

• All 48 unit + integration tests pass (PYTHONPATH=src python -m unittest discover --start-directory src)
  • 34 pre-existing — bit-identical LP output, schema regression-checked
  • 4 new: FlowFiles polymorphic shape (legacy list, dict no-override, dict with override, sibling defaults)
  • 10 new: parse-time fan-out (variant suffix, expansion, deep-copy isolation, declaration-order preservation, legacy no-op)
• Local install via pip install -e . exercised on a real HMCE scenario; output matched the pre-2.2.0 baseline once a stale market-data refresh was identified as the apparent diff source.
• Validated the schema cheatsheet in CLAUDE.md against the actual YAML form by parsing test fixtures.

Out of scope (follow-ups, not in this PR)

• Removing the rebuilder's _stage_simulate_yaml_for_skypro() workaround in skypro-service — needs to wait for 2.2.0 on PyPI, then skypro-service bumps the dependency.
• Hashing imbalance-CSV directories in the rebuilder's data_hash (separate skypro-service ergonomic gap surfaced during release-day investigation; market-data refresh silently invalidates LP outputs without staleness detection).
• The optimised single-dispatch / multi-settlement-column variant of multi-finals (deferred — see updated docs/proposals/multi-final-rates.md).

Release plan

After merge:

1. git checkout main && git pull
2. git tag -a v2.2.0 -m "v2.2.0 — multi-and-per-flow rates schema"
3. git push origin v2.2.0
4. Build (rm -rf dist/ && python -m build) and publish (uv publish dist/* --token "$(cat ~/.simt/pypi.token)")
5. Smoke-test pip install --upgrade skypro && skypro --version