UCT/GDAKI: Expose dev matrix via uct_gdaki_dev_matrix_get (#11702)

* UCT/GDAKI: Expose dev matrix via uct_gdaki_dev_matrix_get

Add uct_gdaki_dev_matrix_elem_t typedef and uct_gdaki_dev_matrix_get
declaration to gdaki.h so external transports (e.g. GDP) can look up
the IB-device→CUDA affinity matrix without duplicating the computation.

uct_gdaki_dev_matrix_get owns the UCS_INIT_ONCE singleton; callers share
one matrix allocation from gdaki.so. uct_gdaki_dev_matrix_init becomes
static. uct_gdaki_query_tl_devices is updated to call the new getter.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

* UCT/GDAKI: Expose dev matrix via uct_gdaki_dev_matrix_get - 2

---------

Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>