I had to debug LLDB's implementation of extend-half-to-float and while it's quite efficient
it also suffered from some UB (see rG4b074b49be206306330076b9fa40632ef1960823 )
and figuring out why we multiply with all those undocumented numbers isn't very fun.
This is a rewrite that is (hopefully) more readable and explains where all the numbers actually
come from. It's also less efficient, but as this is only used for dumping floating point numbers
from a target to a stringstream I don't think the few additional instructions do a lot of harm.
clang-tidy: error: 'lldb/Core/DumpDataExtractor.h' file not found [clang-diagnostic-error]
not useful