feat(gfql): add public Plottable.schema accessor for typed-schema introspection

[lmeyerov]

Background

PR #1457 landed public typed-schema (graphistry.schema.NodeType / EdgeType / GraphSchema / EdgeTopology, bind(schema=...), local gfql_validate()). The bound schema is stored on Plottable._gfql_schema (private attribute, set via bind(schema=...)).

There is no public accessor. Consumers (AI synthesizers, graphistrygpt, schema-aware tooling, debugging code) must reach for getattr(g, "_gfql_schema", None) against a private attribute. This is a P0 ergonomics gap surfaced by the AI-synthesizer user-testing exercise.

Goal

Add a public read-only accessor for bound GraphSchema so consumers can introspect without touching private attributes.

Scope

• Add Plottable.schema (or Plottable.get_schema() — pick whichever fits existing conventions) as a read-only property returning the bound GraphSchema or None if no schema is bound.
• Add Plottable.has_schema() -> bool as a convenience predicate.
• Re-export via top-level graphistry namespace if appropriate per the existing accessor patterns.
• Document the new accessor; mark experimental per #1457's experimental marking.
• Anchored regression tests: bound → returns schema; unbound → returns None; has_schema returns matching bool.

Non-scope

• No changes to bind(schema=...) behavior or internal storage.
• No schema mutation API (read-only accessor only).
• No inference (separate #1338 lane).
• No pretty-printing (separate follow-on issue).

Suggested signature

class Plottable:
    @property
    def schema(self) -> Optional[GraphSchema]:
        """Read-only access to bound GraphSchema (or None if unbound).
        
        Experimental per #1457.
        """
        return getattr(self, "_gfql_schema", None)
    
    def has_schema(self) -> bool:
        """True if a GraphSchema is bound on this Plottable."""
        return self._gfql_schema is not None

Acceptance

• g.schema returns the bound GraphSchema instance (or None)
• g.has_schema() returns matching bool
• Public docs reflect the accessor
• Anchored tests cover bound + unbound cases
• Experimental marking preserved
• Compiler-plan surface touched: no

Cross-refs

• Landed in #1457 / #1337
• Surfaced by AI-synthesizer user-testing 2026-05-25 (P0 finding)
• Coordination: pygraphistry2/'s #1338 inference work; this accessor returns whatever schema is bound (declared or inferred)
• Metaissue: #1058, #1046
• Downstream consumer: graphistrygpt/'s plugins/graphistry/tool.py needs this to retire _chain_contains_radial_layout-style brittle string-matching code

Effort

Small (~30 prod LOC). Drop-in fit for any paused pygraphistry lane.

[lmeyerov]

#1632 coordination note from #1338 implementation PR:

#1636

Accessor integration shape:

• g.bind(infer_schema=True) stores a public GraphSchema instance on g._gfql_schema.
• g.bind(schema=declared_schema) continues to store the provided public GraphSchema.
• graphistry.infer_schema(g) and g.infer_schema() return GraphSchema; return_report=True returns (GraphSchema, SchemaInferenceReport).
• The accessor should be able to expose the bound schema instance without reshaping it. If no schema was bound or inferred, return/empty behavior should remain a feat(gfql): add public Plottable.schema accessor for typed-schema introspection #1632 decision.

Important guardrail:

• bind(schema=..., infer_schema=True) is intentionally rejected so the accessor never has to explain an implicit merge from one bind call.

[lmeyerov]

Coordination note for #1485 PR #1638: the returned graph after a successful local schema-effect call carries an updated bound GraphSchema in the same private _gfql_schema slot. The #1632 accessor should be able to expose that post-effect schema without reshaping it. This PR does not add a public accessor and does not expose SchemaEffect.

[lmeyerov]

PR #1637 is ready for manual review: #1637

Implemented experimental read-only g.schema + g.has_schema() over the existing bind(schema=...) storage path. Bound/unbound/read-only tests, docs, changelog, review-skill convergence, and full PR CI are green. No compiler-plan surface touched; no DGX run because no GPU/cuDF/dataframe runtime behavior changed.