Revert "Remove outdated "output directory name" option."

[torgil]

This reverts commit af0e20f.

[torgil]

See discussion in #14236

[gregestren]

@sdtwigg triage / close?

[sdtwigg]

Sorry for the delay here, an initial skim of the title missed the effective double negative so I was reading it as just "remove outdated output directory name option" and thus thought it was the initial PR for a change that went through and thus was an accidental artifact.

That said, we broadly cannot trust the user to override and set the output directory name. There is too much nuance in making sure action conflicts are avoided, especially as the situation gets more complex with deeper build graphs and more transitions. This is actually why I am doing a heavy push to have output directory be computed directly by looking at just the current set of options (and maybe comparing to a baseline) rather than relying on the transition or transition infrastructure to do the right thing immediately.

Instead, I think we should fix the underlying issue by more judicious use of EXPLICIT_IN_OUTPUT_PATH on options as well as named configurations/transitions and better unifying behaviors with the exec transition. Given all of the work in this space since the initial underlying issue was files, could you re-evaluate if it is still a problem (especially when using --experimental_output_directory_naming_scheme=diff_against_dynamic_baseline) then re-submit a fresh issue?