Add GNNGraphsEnzymeCoreExt marking scaled_laplacian inactive for Enzyme

[Parvm1102]

scaled_laplacian is already @non_differentiable for ChainRules, but Enzyme does not
read those declarations, so it differentiated the Krylov eigensolve inside it and failed
with an IllegalTypeAnalysisException in SparseArrays. The Laplacian depends only on the
graph, never on the node features, so treating it as constant loses no gradient.

This adds a small EnzymeCore package extension (same pattern as the Mooncake extension)
with a single EnzymeRules.inactive declaration, plus a test.

Unblocks ChebConv, GConvGRUCell and GConvLSTMCell under Enzyme; GConvGRU/GConvLSTM
additionally need a separate GNNRecurrence fix.

[Parvm1102]

Please review this @CarloLucibello