Graph algorithm only connects episodes by shared concepts

[CalebisGross]

Problem

internal/api/routes/graph.go lines 113-148: The episode graph uses O(n^2) concept matching with a minimum threshold of 2 shared concepts. This produces sparse, disconnected graphs for most datasets.

Missing edge types:

• Temporal proximity (events close in time)
• Causal relationships from the associations table
• Association-based edges (the memories view at lines 162-229 uses these, but episodes view does not)

Impact

The graph visualization is mostly disconnected islands with few meaningful connections.

Fix

Use actual association data from the store, add temporal proximity edges, and consider using the memories view as default.

[CalebisGross]

Fixed in PR #44 — episode graph now uses association edges, temporal proximity edges, and improved shared concept edges with deduplication.