ggml : reduce CPU overhead in meta backend (llama/22041)

ggml : reduce CPU overhead in meta backend (llama/22041)

* cache subgraph splits when cgraph is unchanged

Skip per-call subgraph construction in ggml_backend_meta_graph_compute when the same ggml_cgraph is used consecutively.

Assign uid to every sub-graph so that CUDA's fast uid check path hits too.

* Address review comments

* Keep the scope as is

* Rename last_uid and last_n_subgraphs field. Remove last_max_tmp_size field. Refactor code.

* Address review comments

* Update ggml/src/ggml-backend-meta.cpp

Co-authored-by: Johannes Gäßler <redacted>
* Update ggml/src/ggml-backend-meta.cpp

Co-authored-by: Johannes Gäßler <redacted>
---------

Co-authored-by: Johannes Gäßler <redacted>