potentially accelerate trace/symbol recovery by reusing buffers

There is a potential issue in `do_resolve`, where we rebuild the return-value buffer each time on the stack: https://github.com/rust-lang/backtrace-rs/blob/dccdb4d50f1ccd0cf08d8a83a0344e694b65f97e/src/symbolize/dbghelp.rs#L220-L232 This sort of thing might happen elsewhere in the code, too. We're not expected to be the fastest thing but if we find free performance lying on the floor, well, why not?