perf: cranelift-srcgen: optimize Formatter buffering

[dishmaker]

This reduces run-time of cranelift-codegen (build-script) (run):

• -0.4% on debug
• -3.1% on release ¹

by changing:

• lines: Vec<String> to
• lines: String

so that

file.write_all(self.lines.as_bytes())?;

is called once.

Depends on #14092

Footnotes

1. build script won't be in opt-release mode, because of [profile.release.build-override] opt-level = 0
   https://github.com/bytecodealliance/wasmtime/blob/a48196189a5e3b5e5f288f776062fb6e7189e43f/Cargo.toml#L177-L178 ↩