Split up encoding settings & defaults to differentiate local vs subprocess

[bitprophet]

Spinoff of #350, tl;dr we still have a single encoding option/kwarg and a single default_encoding() Runner method. However, it's plausible (especially for other subclasses like remote ones) for these to not be similar environments, e.g. the subprocess is using UTF-8 but one's local env is C or ASCII or whatnot.

Ideally we'll split things up into e.g. our_encoding/ subprocess_encoding (to match the other IO-boundary methods like write_our_output/read_proc_stdout). There are TODOs in the code where this may want to occur.