Allow capturing of stderr as part of custom_target

[mitchgrout]

Certain tools may emit useful information into the stderr stream, which we may want to capture as a custom target. Currently we can capture stdout using custom_target(..., capture: true), however there doesn't appear to be a way to get at the stderr. It's relatively trivial to write an external script to manage swapping streams, like given below:

#!/usr/bin/env ruby
exec(*ARGV, :out => :err, :err => :out)

In my opinion it would be more appropriate to hoist this behaviour into meson, since we already facilitate capturing of stdout. My proposal would be to change the type of capture from bool to bool | dict[str] while deprecating the bool variant. The dict[str] variant would be of the form

capture: { 'stdout': str, 'stderr': str }

where the values in the dict would represent which files we will capture the specified stream into. If a stream is not specified as a key, then it is not captured. The default value for capture would be an empty dict, meaning neither stdout nor stderr are captured by default. As an example, the following statements would be equivalent:

# Current style
custom_target(command: [generate, '@INPUT@'], input: ..., output: ..., capture: true)

# Proposed style
custom_target(command: [generate, '@INPUT@'], input: ..., output: ..., capture: { 'stdout': '@OUTPUT@' })

However, with the new dict[str] variant we would be capable of capturing from tools which emit to stderr rather than stdout, e.g.

custom_target(command: [generate, '@INPUT@'], input: ..., output: ..., capture: { 'stderr': '@OUTPUT@' })

and would also support capturing both streams simultaneously

custom_target(command: [generate, '@INPUT@'], input: ..., output: [..., ...], capture: { 'stdout': '@OUTPUT0@', 'stderr': '@OUTPUT1@' })

---

This request primarily originates from trying to integrate linting into my project, in which I was using a generated executable to manage swapping the streams (which lead to discovering #12040)

[mitchgrout]

One cleaner solution I suppose could be to instead hoist this into find_program itself, e.g.

find_program(...,, swap_streams: true)