epic(correlation): entity-resolution-first correlation improvements

## Goal

Improve correlation coverage by making canonical entity identity a controlled, measurable join key while preserving existing cards, trend continuity, and no-deploy rollback.

## Committed scope

- Phase 1a: exact alias-tier entity clustering with theme fallback across the browser and Railway seeder.
- Phase 1b: fuzzy resolution only if the residual-miss gate passes and its candidate-generation and performance-budget decisions are resolved.
- Staged runtime modes: `legacy`, `exact`, and `fuzzy`; missing or failed configuration remains `legacy`.

## Completion criteria

- A representative baseline and labeled acceptance benchmarks establish current misses and user impact.
- Exact entity clustering preserves existing emitted cards and passes precision, salience, latency, and browser/seeder parity gates.
- Exact and fuzzy behavior can be activated and rolled back independently without a deployment.
- Phase 1a is activated only after explicit operational approval and its acceptance gates pass.
- Phase 1b is either rejected by its numeric gate or implemented, statistically validated, and separately approved for activation.

## Not completion scope

Reciprocal-rank-fusion ranking, hybrid semantic/structural similarity, causal chains, and conflict detection remain separately linked discovery work. They do not block this epic.

## Delivery issues

- #5982 — Baseline and acceptance dataset
- #5983 — Staged runtime correlation modes
- #5984 — Exact entity clustering in the browser
- #5986 — Railway seeder parity
- #5987 — Validate and activate exact mode
- #5988 — Decide Phase 1b eligibility and architecture
- #5989 — Implement conditional fuzzy resolution
- #5990 — Tune, validate, and activate fuzzy mode

The initial execution frontier is #5982 and #5983. Activation in #5987 and #5990 requires separate operational approval.

## Related discovery

- #5991 — Reciprocal-rank-fusion ranking
- #5992 — Hybrid semantic and structural similarity
- #5993 — Decay-weighted causal chains
- #5994 — Cross-source conflict detection

These issues are discovery-only and do not block epic completion.

Start with delivery issues #5982 and #5983, then read the linked Phase 1a, parity, validation, and activation issues in sequence. Use the completion criteria as the acceptance checklist: baseline metrics, exact clustering with browser/seeder parity, independently staged rollback, and explicit operational approval before activation.

Written by the indexing model from the issue text.