Signed integer overflow being UB enables compiler optimizations. Most famously, when compiling

  for (int i = 0; i < bound; i++)

in an LP64 environment (i.e., one where the native machine word is 64 bits but int is 32 bits, as is the case in most modern compilers targeting most modern architectures), it allows i to be represented by a native machine word, increasing performance.

You can, and people constantly do, argue about whether this is worth it, but I suspect it's the primary reason why a lot of the committee didn't want to change this behavior.