Add IndexCache support for GLM5 DSA (#45424)
* Add IndexCache support for GLM5 DSA
* Refactor: Make IndexCache layer scheduling explicit in Config
Moves index_topk_pattern generation from Attention.__init__ to
Config.__post_init__ as suggested. Layers now simply check
`config.index_topk_pattern[layer_idx]` instead of computing
skip conditions, matching the mlp_layer_types pattern for
consistent explicit configuration.
* fix
* oof, typo
* remove the exception as its now hidden behind kwargs for BC
---------
Co-authored-by: vasqu <antonprogamer@gmail.com>
