Bad codegen when not using range in switch prong

Zig Version

0.14.0-dev.2273+73dcd1914

Steps to Reproduce and Observed Behavior

export fn a(x: u8) bool {
    return switch (x) {
        99, 1, 2 => true,
        else => false,
    };
}

export fn b(x: u8) bool {
    return switch (x) {
        99, 1...2 => true,
        else => false,
    };
}

The codegen for these two functions that clearly do the exact same thing differs, with the version not using ranges probably having inferior codegen:

a:
        mov     al, 1
        lea     ecx, [rdi - 1]
        cmp     ecx, 2
        jb      .LBB0_3
        cmp     edi, 99
        je      .LBB0_3
        xor     eax, eax
.LBB0_3:
        ret

b:
        cmp     edi, 99
        sete    cl
        lea     eax, [rdi - 1]
        cmp     al, 2
        setb    al
        or      al, cl
        ret

https://zig.godbolt.org/z/fhWK7vs9K

Expected Behavior

Both versions should compile down to the same optimal codegen.
This applies to multiple architectures in both ReleaseFast and ReleaseSmall.

I feel like once this is fixed #22094 could be reverted if it's merged and then for (whitespace) could be changed to inline for (whitespace) to yield equally good codegen as with the changes of that PR.