Fix PV→PQ bus type conversion when bus rows are not in ID order

[yemine0x01]

• Root cause: gen_count is produced by np.bincount over gens_on[:, GEN_BUS] (0-based bus ID), so it is indexed by bus ID. The mask (self.buses[:, BUS_TYPE] == PV) & (gen_count == 0) was comparing element-wise against gen_count in row order. When pypowsybl exports buses in non-sorted order (MATPOWER, PSS/E, XIIDM formats), bus row ≠ bus ID, so the wrong buses were promoted from PV to PQ.
• Fix: Index gen_count with bus_ids = self.buses[:, BUS_I].astype(int) so the comparison aligns with the bus row being evaluated.

[albanpuech]

Looks good.
However, here are many parts of the code in datakit and in graphkit that expect the buses to be sorted. Just to give one example: https://github.com/gridfm/gridfm-graphkit/blob/a5e6449f5fd5f77211ba22483ef8e7f3f14c6120/gridfm_graphkit/datasets/powergrid_hetero_dataset.py#L162

[yemine0x01]

Thanks for the review, @albanpuech. Just to clarify, do we strictly need the buses to be ordered by ID? For context, these changes don't remove the existing bus ordering; they simply ensure the logic still functions correctly even if the buses happen to be unordered.

[albanpuech]

Thanks for the review, @albanpuech. Just to clarify, do we strictly need the buses to be ordered by ID? For context, these changes don't remove the existing bus ordering; they simply ensure the logic still functions correctly even if the buses happen to be unordered.

There are many places in the code that assume that, yes.
e.g.

gridfm-datakit/gridfm_datakit/process/process_network.py

Line 650 in 75dfdcf

Pg_bus = np.bincount(gen_bus, weights=pg_gen, minlength=n_buses)

or https://github.com/gridfm/gridfm-graphkit/blob/a5e6449f5fd5f77211ba22483ef8e7f3f14c6120/gridfm_graphkit/datasets/powergrid_hetero_dataset.py#L162

We would need to check everything and list where code needs to be changed