Feature request: make graph queries context-aware across languages

[TheFedaikin]

What we want

We want graphify queries to distinguish between edge kinds that are currently traversed together.

Examples:

• calls vs type/declaration references
• imports vs executable relationships
• field references vs other symbol mentions

A query like who calls X? should prefer call-heavy results instead of mixing calls, imports, and declaration-style references.

Why

This improves query precision in larger or noisier graphs and makes edge disambiguation a graph-level, cross-language capability.

Model boundary:

• relation = semantic edge type
• context = syntactic disambiguation where needed

Proposal

Add a cross-language context layer for edges where it improves query quality, and use it during query traversal.

Initial contexts:

• call
• import
• field

Queries should support:

• explicit filtering
• small heuristic inference when the query clearly implies a context

Examples:

• graphify query "extract" --context call
• who calls extract -> infer call when no explicit filter is provided

Scope

• add/standardize edge context where supported
• support context-aware filtering in:
  • CLI graphify query
  • MCP query_graph
• show active context filtering in query output

Non-goals

• do not move inherits / implements into context
• do not replace relation with context
• do not build broad NLP intent parsing
• do not treat this as a fix for general explain / path disambiguation

References #569

This is related to #569, but narrower in scope.

• #569 is mainly about resolving the right node or candidate set in large, noisy graphs
• this issue is about traversing the right edge kinds once the node is found

In short:

• #569 = node resolution / disambiguation
• this issue = edge-context-aware traversal

Done when

• queries can explicitly filter by context
• obvious query phrasing can infer context when no filter is passed
• output makes the active context visible
• extraction/tests cover the touched languages