[RFC] allow user actions to request direct access to the terminal

Dune systematically buffers the output of commands so that their output is not mixed up during parallel builds. However, this is not ideal for interactive commands that require direct access to the terminal.

Currently, the only workaround is to run dune with -j 1 --no-buffer which is not ideal since it completely disables parallelism and also require a specific dune invocation.

Another idea would be to let actions declare that they want direct access to the terminal.

Proposal

Add a new action (using_terminal ...) to introduce an action that requires the terminal. For instance:

(rule
 (alias runtest)
  (action (using_terminal (run ./prog.exe))))

Only one such action can run at a time, and while such an action is running messages emitted by dune would be put in a queue and rendered once the action has finished.

Start by reviewing Dune's action execution and output-buffering behavior, then compare it with the proposed (using_terminal ...) action. Done means interactive actions can access the terminal directly, only one such action runs at a time, and Dune messages are queued until it finishes.