[llvm] Fix to account for multiple ISA enumeration (PR #118676)

[Shilei Tian via llvm-commits]

shiltian wrote:

> Good short-term fix, long term we just need to handle multiple of these and push them back. Then we can scan all the ISAs from least generic to most generic through some complicated lookup table.

This is not a short term fix. The iteration always starts from specific to generic, so there must be only one match. It is just code error not stopping it on the first match.

https://github.com/llvm/llvm-project/pull/118676