FAT v2 P3: generalized multi-source persistence and package bundle

[masarray]

Stacked on proven FAT v2 P2

Base branch: feat/fat-v2-selection-authority
Base SHA: e5c081c5eb494d43b7d26623b9c528dab20ff219
P3 proven head: cb785bb06d95f838e8bafad5e1978699bc56e117

P3 is intentionally limited to FAT source identity, workspace persistence, and portable package handling. It does not implement SCL -> FAT row projection; that remains P4.

P3 contract

• generalized FAT source descriptors for workbook and scl, each with exact SHA-256, deterministic source ID, filename and length
• deterministic, order-independent source-set fingerprint so selecting the same SCL files in a different order resolves to the same FAT project identity
• one legacy workbook preserves its historical storage fingerprint (SourceWorkbookSha256) so existing local snapshot directories remain discoverable after P3
• OpenSourcesAsync stages one or many source files into the local workspace using deterministic collision-safe names
• snapshots persist sources[] plus sourceSetSha256
• snapshot bootstrap matches the complete source set; pre-P3 snapshots without sourceSetSha256 still match by the legacy workbook SHA
• .arsas packages carry and verify the complete source bundle, including every per-source SHA/length and the aggregate source-set fingerprint
• package import accepts both the new source-bundle manifest and legacy singular-workbook manifests
• SCL-only projects never synthesize a fake workbook; SourceWorkbookPath is empty when no workbook exists
• SCL-only .arsas export still generates the native PDF evidence report but deliberately omits workbook-derived IO-FAT-Results.xlsx
• workbook-backed projects retain the existing Excel result export and legacy SourceWorkbookPath compatibility accessor
• no SCL XML parser or SCL -> FAT signal projection was added in P3

Regression coverage added

• source-set fingerprint is order-independent
• legacy one-workbook storage path remains stable
• two-SCL workspace restores operator progress even when the source selection order is reversed
• multi-SCL .arsas round-trip carries all sources and restores them without a fake workbook/Excel artifact
• tampered SCL source content is rejected by package validation
• existing workbook persistence/package regressions remain green

Scope audit

Compared with proven P2 head, this branch is 9 commits ahead and 0 behind. Exactly 7 FAT source/persistence/test files changed:

• Models/IoTesting/IoFatSourceModels.cs
• Models/IoTesting/IoTestModels.cs
• Services/IoTesting/IoFatProjectPackageService.cs
• Services/IoTesting/IoFatSourceWorkspaceService.cs
• Services/IoTesting/IoTestProjectPersistenceService.cs
• Services/IoTesting/IoTestWorkspaceBootstrapService.cs
• tests/ARSAS.Tests/IoFatSourceBundlePersistenceTests.cs

No ARIEC engine pin, report planner, IEC 61850 protocol/control path, SCL authority/parser, or P2 rolling-capture implementation changed.

Exact-head evidence

Exact P3 head cb785bb06d95f838e8bafad5e1978699bc56e117:

• Validate IO List Testing #454 — SUCCESS
  • focused project restore — SUCCESS
  • focused IO compiler — SUCCESS
  • complete IO regression suite — SUCCESS
• Validate SV evidence bundles #623 — SUCCESS
  • focused compiler — SUCCESS
  • SV evidence regression suite — SUCCESS
• Build ARSAS #1511 — SUCCESS
  • source/license boundaries — SUCCESS
  • premium UX / GOOSE / SMV / SAS invariants — SUCCESS
  • immutable ARIEC API verification — SUCCESS
  • solution restore/build — SUCCESS
  • full ARSAS application regression suite — SUCCESS
  • portable single-EXE publish — SUCCESS
  • portable single-EXE smoke test — SUCCESS

ARIEC61850 remains on the proven pinned revision. The branch is intentionally left unmerged and this PR remains draft for stacked-phase review.

[masarray]

Closing as superseded by consolidated FAT PR #236. The P3 source-bundle/persistence work is already in the active FAT branch ancestry; this stacked PR no longer needs an independent merge path.