Bump default flash-attn2 hub kernel version to v3 (#47863)

* Bump default flash-attn2 hub kernel version to v3

The default version for attention kernels loaded from the Hub was hardcoded
to v1. The v1 branch of `kernels-community/flash-attn2` only ships builds up
to Torch 2.12, so loading it on Torch 2.13 fails with no matching build. v3
ships the Torch stable ABI (CUDA) and Torch 2.13 builds, including XPU.

Introduce a per-repo version pin table instead of bumping the global default,
since other attention kernel repos (vllm-flash-attn3, msa, flash-mla, ...) are
still on v1.

* fix attn version check

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>

---------

Signed-off-by: jiqing-feng <jiqing.feng@intel.com>
Co-authored-by: Ilyas Moutawwakil <57442720+IlyasMoutawwakil@users.noreply.github.com>