perf: cranelift-codegen-meta: less llvm-lines: strings.join(" ") and iterators - #14127

-2.2% llvm lines

• added DisplayJoined
• simplified iterator combinators: .chain(...)
• use format_args! in fmt.add_block calls

This PR might be controversial, because anywhere in the code just one:

• strings.join(" ") instead of
• strings.display_join(" ")

adds back the 1254 llvm-lines that I managed to remove 😐
(but that's still a -1.5% win).