model: prepare source-neutral DataSet capability index and parity fingerprints

[masarray]

P1 stacked on P0

Base: feature/discovery-static-workflow-parity-p0

This is intentionally stacked on the P0 discovery/static-workflow candidate. P0 remains the field-test boundary; this PR does not change MMS/report activation traffic.

Goal

Prepare one source-neutral, immutable DataSet capability view for both Open SCL and Smart Discovery so downstream UI/report planning can consume the same precomputed facts.

Initial P1 slice

• add ModelGeneration to invalidate stale model-derived capability work;
• build the capability index off the UI path with cancellation of superseded generations;
• index static DataSets, canonical ordered members and configured RCB bindings;
• compute deterministic SHA-256 fingerprints for DataSet structure and stable RCB configuration;
• exclude mutable runtime state such as RptEna/reservation from parity fingerprints;
• warm the same capability worker after Open SCL and after discovery;
• let the Static DataSet chooser consume the prepared index when ready, with direct-model fallback;
• add parity comparison helpers and unit/regression tests.

Safety boundary

This P1 slice does not:

• issue MMS reads;
• browse DataSet directories;
• enable RCBs;
• create dynamic DataSets;
• add polling;
• change P0 report activation;
• publish or merge any release.

Intended parity contract

Discover -> capability fingerprint and Open SCL -> capability fingerprint should match for:

• DataSet identity;
• exact member index/order;
• member reference / FC / MMS reference;
• configured RCB -> DataSet binding;
• stable RCB reporting configuration.

Runtime-mutating RCB state is evidence, not configuration parity, and is deliberately excluded.

[masarray]

P1 initial slice is active on top of the P0 field-test branch.

Implemented so far:

• immutable source-neutral DataSet capability index;
• model-generation invalidation so stale background work cannot overwrite a newer model;
• cancellation of superseded capability workers with worker-owned token disposal;
• deterministic SHA-256 fingerprints for DataSet member order/reference/FC/MMS reference;
• deterministic fingerprints for stable configured RCB -> DataSet bindings;
• runtime-mutating RCB state is excluded from parity fingerprints;
• the same worker is warmed after Open SCL, live discovery, fast model connect and card-scoped IED actions;
• chooser consumes the prepared capability index when available and falls back directly to the canonical model when preparation is still running;
• no MMS read, DataSet browse, RCB write, dynamic DataSet creation or process polling is performed by this worker.

Current CI evidence on this stacked branch:

• Merge Execution Guard: success;
• ARSAS regression tests in that guard: success;
• Field Capture Build has compiled the P1 source successfully and is continuing packaging;
• R7 Build has compiled the P1 source successfully and is continuing canonical-export tests.

This PR intentionally remains Draft. Do not use P1 as the P0 physical-test binary; P0 PR #339 remains the isolated field-test authority.