P4 #144: targeted RCB hot path and association-scoped static DataSet evidence

[masarray]

Stacked performance slice for #144. Base is fix/144-smart-interop-p0 (PR #145) so this work stays isolated from the active semantic/export and ARSAS packaging lanes.

Goal

Make runtime acquisition proportional to the selected signals/RCBs, not to the size of the IED.

Smart means less network work with stronger context, not more probing.

Changes

• add exact TargetReportControlReferences to availability options;
• broad diagnostic mode remains unchanged when no targets are supplied;
• targeted mode never broadens a missing exact live RCB target to sibling/unrelated RCBs;
• extract a pure target selector so family resolution/planning and network probing remain separate;
• targeted availability uses a minimal structure-first probe:
  • one complete RCB structure read first;
  • exact RptEna leaf only if still unresolved;
  • DatSet leaf only if structure did not prove the binding;
  • Resv/ResvTms only if still unresolved;
  • Owner only when the live RCB exposes it;
• broad availability retains the full diagnostic attribute probe;
• successful non-deletable DataSet directories are cached for the current MMS association only;
• dynamic/deletable or unknown-mutability DataSets are never cached;
• cache clears on every accepted MMS association;
• availability diagnostics expose inventory/probed target counts, RCB logical read budget, DataSet network reads and cache hits.

Anti-naive boundaries

• no vendor-name branching;
• no global reduction of availability safety;
• no cache of RptEna/Owner/Resv/ResvTms;
• no cache across association boundaries;
• no family-name guessing in the network probe layer;
• no fixed sleeps;
• no full-RCB inventory scan when exact live targets are supplied.

Regression

Pure selector tests prove:

• empty target set preserves broad mode;
• exact targets select only requested live RCBs;
• missing target is not broadened to a sibling RCB;
• hard MaxReportControls still applies.

Consumer plan

ARSAS #446 should eventually pass only concrete live RCB references produced by canonical static coverage + RCB family resolution. The ARSAS engine lock is intentionally untouched here because active PR #425 currently owns that integration surface.

Canonical-to-live target resolution

This slice now also keeps target selection inside ARIEC61850 rather than forcing ARSAS to reproduce IEC 61850 semantics:

• CanonicalLiveModelAdapter.FromSclWorkspace(...) gives Open-SCL a source-correct canonical ingress instead of relabeling the SCL design model as live discovery;
• MmsCanonicalStaticRcbTargetResolver resolves canonical covered DataSets to concrete live RCB references using only exact configured-reference identity or exact live DatSet binding + BRCB/URCB family;
• indexed/sibling RCB names are never fabricated;
• unresolved segments remain unresolved and may explicitly fall back to a broader diagnostic search;
• tests lock exact DataSet targeting, BRCB/URCB family isolation, exact-reference fallback, and no name-family guessing.