fix: validate disconnected nodes and exclude input nodes from outputs

[Rkx-01]

Problem

Closes #244
Supersedes #245

Two bugs in model_generation.py silently broke model generation:

• Disconnected nodes caused a KeyError crash — BFS never visits
  unconnected layers but the code still tries to access them in
  keras_tensors

• Input nodes were misidentified as outputs — custominput nodes
  with no outgoing edges were included in output_ids, producing
  malformed Keras models silently

Fix

• Added validation after BFS loop that raises a clear ValueError
  listing exactly which nodes are disconnected
• Fixed output_ids filter to exclude custominput nodes

Testing

• Disconnected canvas nodes → clear error message shown ✅
• Connected canvas → model generates correctly ✅
• Iris sample project still works perfectly ✅

Note

Fresh PR as requested by @ivantha from #245.
Rebased on latest main, single clean commit, all checks passing.

[Rkx-01]

Hi @ivantha - fresh PR superseding #245.
Rebased on latest main, single clean commit, all checks passing.
Ready to merge!