Create test for selecting derivation outputs in "installables" with .

[Ericson2314]

Accessing an output in an installable with . (foobarbaz.<output>) now has the original behavior again:

• Originally: nix build nixpkgs#libxml2.dev will build the dev output
• After nix: Respect meta.outputsToInstall, and use all outputs by default #6426: nix build nixpkgs#libxml2.dev wil build all "outputs to install", not just `dev!
• After Respect the outputSpecified attribute #6589: Once again, nix build nixpkgs#libxml2.dev will build the dev output

However I don't think anything tests that behavior yet: see how #6589 didn't change any tests or add any new ones. We should add a new one.

[Ericson2314]

Perhaps this would be a good task to onboard new people?

I suspect #6589 might currently just being leveraging a Nixpkgs convention, but if so this is easy to fix in the derivation primop wrapper. I could easily be wrong though, this is just from a quick skim, but I wanted to flag this so that anyone taking this up didn't get stuck.

In particular:

 $ git grep outputSpecified
src/libexpr/eval.cc:    , sOutputSpecified(symbols.create("outputSpecified"))

The literal name of the attribute only occurs in the symbol.

 $ git grep sOutputSpecified
src/libcmd/installables.cc:    if (auto aOutputSpecified = attr->maybeGetAttr(state->sOutputSpecified)) {
src/libexpr/eval.cc:    , sOutputSpecified(symbols.create("outputSpecified"))
src/libexpr/eval.hh:        sOutputSpecified;
src/libexpr/get-drvs.cc:    if (attrs && (i = attrs->find(state->sOutputSpecified)) != attrs->end() && state->forceBool(*i->value, i->pos)) {

sOutputSpecified seems to be read but never set.

 $ git -C path/to/nixpkgs grep 'outputSpecified ='
lib/customisation.nix:            outputSpecified = true
...

however it is set in Nixpkgs.