test(episodes): add v0.7 continuous-background oracle

[stacknil]

Summary

• add a sanitized v0.7 research fixture with two five-event dense peaks connected by nine-minute background gaps
• record the exact v0.6 baseline: one activity segment, one selected episode, and deterministic first-maximum tie-breaking
• define a Draft 2020-12 candidate-oracle format for future comparison without implementing a candidate detector

Why

The episode-semantics ADR identifies continuous low-density background as the narrow research boundary where adjacent-gap segmentation can hide a second dense peak. This PR makes that behavior measurable before any v0.7 segmentation policy is proposed.

How to validate

cmake -S . -B build
cmake --build build --parallel 1
ctest --test-dir build -C Debug --output-on-failure

Additional review checks performed locally:

• baseline reconstruction: 15 events, maximum adjacent gap 540 seconds, one baseline segment, selected evidence line:1 through line:5
• deterministic finding identity: finding:brute_force:584fd14b544a7959
• Draft 2020-12 meta-schema validation plus a complete two-episode candidate example
• git diff --check and privacy scan using only the RFC 5737 placeholder 203.0.113.77

Design decision

Keep the v0.6 detector and loglens.report.v3 unchanged. The fixture records current behavior, while the candidate oracle captures experiment provenance, windows, episode selections, event decisions, and baseline comparison.

Main risk

The oracle shape could constrain future experiments too early. It is therefore explicitly research-only and describes comparison evidence rather than selecting a segmentation algorithm or stable report contract.

Compatibility impact

None. This PR adds three JSON test artifacts only; it changes no detector logic, parser behavior, configuration, CLI output, or report schema.

Rollback path

Revert commit 592ce392258e2fd4222ea0bc42a12dff8ab89cef; no migration or generated-output cleanup is required.

[stacknil]

Final diff review completed after green checks.

• Scope: exactly three research-only JSON artifacts; no detector, parser, config, workflow, CLI, or report changes.
• Design decision: preserve the v0.6 adjacent-gap/best-window baseline and separate baseline evidence from any future candidate output.
• Main risk: the oracle could overconstrain future research; it is explicitly versioned and research-only, with no candidate algorithm selected.
• Compatibility impact: none for loglens.report.v3 or current runtime behavior.
• Rollback path: revert 592ce392258e2fd4222ea0bc42a12dff8ab89cef; no migration or cleanup is required.
• Validation: local artifact reconstruction and Draft 2020-12 validation passed; CTest passed 6/6; all 11 remote check runs concluded success.

No blocking findings.