Vulkan: flash attention for DeepSeek models (#584)

* vulkan: support mixed/deepseekR1 FA head sizes (#14509)

* vulkan: better parameterize FA by head sizes

* vulkan: support mixed/deepseekR1 FA head sizes

* Fix the FA cherry-pick

---------

Co-authored-by: Jeff Bolz <[email protected]>
Co-authored-by: Iwan Kawrakow <[email protected]>

Author: 3 people