stage1: `@floatToInt` safety check causes LLVM segfault on aarch64-linux-none

[topolarity]

Zig Version

0.10.0-dev.1740+23715dcf6

Steps to Reproduce

Execute the following test with -target aarch64-linux-none or arm-linux-none:

test {
    var y = @as(f80, 2.5);
    try testing.expect(@floatToInt(u80, y) != 0);
}

Expected Behavior

Compilation and test should both succeed.

Actual Behavior

Segfault occurs in LLVMConstReal when adding the safety check for the floatToInt operation (codegen.cpp:3854):

Segmentation fault at address 0x16b0
???:?:?: 0x63d64c9 in ??? (???)
[1]    6592 IOT instruction  ./build/zig test testme.zig -target aarch64-linux-none --zig-lib-dir lib

I believe the problem is that we are lowering a FP constant for a value that we actually lower as an integer on ARM (and the safety check therefore also needs to be implemented in soft ops).