[TS PBT] Persist direct campaign results with existing evidence

[CaelmBleidd]

Direct campaign results currently remain in memory. Add PropertyCampaignRunner.runArtifact, followed by artifact.write(path) and PropertyCampaignArtifact.read(path), to retain and inspect a completed run without executing its engines again.

The typed artifact preserves the original PropertyDefinition, the declared PBT controls and validation reserve, original-property target role, and the existing concrete, symbolic and confirmation results. Their replay/reduction evidence, coverage, observations, candidates, diagnostics and stop phase remain intact. Existing symbolic result/capability types are already portable, so this change makes them serializable rather than introducing parallel snapshot models. The existing UTF-16-safe JSON boundary preserves JavaScript special values and lone surrogates in UTF-8 files. Creation and reading reject mixed property IDs, including IDs inside concrete, symbolic and confirmation results.

Source/tool/runtime identities are retained only where the existing property or result provenance records them. Constructor machine options, explicit tsconfig and runtime paths are not attested by this artifact; it is not a complete environment snapshot. Campaign execution and its budget semantics remain those of #540. CLI integration is #545 and the one-way campaign handoff is #355. Cross-round sessions and repeated random PBT generation are outside the current scope; this artifact delivery does not by itself close #354.

Stacked on #540 (caelmbleidd/pbt-direct-campaign-controls) at 71fdab8feabe6b16ccd46a33e1589ee1de320992, including the shared process-shutdown repair from #506.

Validation: full module checks passed after updating the shared transport: 195 PBT Kotlin tests, 138 FastCheck Kotlin tests and 186 Node tests, without failures or skips. These include 6 new real run → file → read scenarios, 13 direct-campaign regressions and 18 symbolic-search regressions. The new scenarios cover native violations with actual coverage and observations, native confirmation and reduction, unsupported replay, confirmation timeout with a retained candidate, special values, and rejection of mixed identities. Strict Detekt main/test checks for both changed modules and git diff --check passed.