fix: require source/target roles in binary relation_members validation

[dex0shubham]

Problem

For binary graph templates, _check_relation_members only verifies that each relation_members dict value names a valid edge field — it accepts arbitrary keys:

for role, field_name in members.items():
    if field_name not in relation_fields:
        diags.append(Diagnostic(HE_T003, "error", ...))

But the runtime extractor (parsers/identifiers.py) reads the roles by name:

source_field = members["source"]
target_field = members["target"]

So a template like relation_members: {from: source, to: target} (whose values are valid edge fields) passes validation and then raises KeyError: 'source' at extraction time — precisely the failure the validator is meant to catch ahead of time.

Fix

In the binary-graph branch, require the relation_members dict to define both a source and a target role, emitting an HE-T003 error for each missing one.

Tests

tests/template_engine/test_validator.py: a graph template using non-standard role keys (from/to) with valid field values is now rejected with source/target diagnostics. The full validator suite — including TestPresetBootstrap, which validates every bundled preset clean — still passes (all presets already use source/target). The new test fails on the pre-fix code.

ruff check/ruff format --check on hyperextract clean.

[yifanfeng97]

Merged — thanks! 🎯

Verified locally: the runtime extractor does read members["source"]/members["target"] directly, so a template with non-standard role keys (e.g. from:/to:) passed validation but raised KeyError at extraction time. The validator now rejects this up front with a clear HE_T003 diagnostic, and the regression test covers it.

This will ship in the next release.