Possible panic safety issue in StderrForwarder

[daira]

When auditing changes to cc, I found a theoretical panic safety issue at 227b770#diff-fbc116db4f047e6e29dac1fc3c5a5f1f759060e9558aff5707624841b57c5258R123-R128 . As an optimization, the code extends the length of the buffer to its capacity (which will cause it to include uninitialized data), and relies on it being set back to contain only initialized data in all code paths. If stderr.read panics and that panic is caught by a caller of forward_available, then the inner buffer of StderrForwarder will still contain uninitialized data. I don't think it is likely that this can be triggered maliciously, hence the public issue.

[NobodyXu]

Thanks, I think replacing it with RAII should fix it.
Would probably submit a PR later, but if you already have one, I'm happy to merge it