Manager.load_profile: do not load default profile if other already loaded

[sphuber]

Fixes #5381

The Manager.load_profile method was loading the default profile if no
explicit profile name was specified, even if another profile was already
loaded. This is violating the note in the docstring that if a profile is
already loaded, not specifying an explicit profile to be loaded should
be a no-op.

This bug was manifesting itself in verdi where the --profile option
was effectively ignored if a profile other than the default was
specified for a command that uses the load_dbenv decorator. This is
because the ProfileParamType would correctly load the profile
specified by the option, but then the load_dbenv decorator would call
aiida.cmdline.utils.decorators.load_backend_if_not_loaded which would
call Manager.load_profile without specifying a profile, and so the
default profile would be loaded.

[sphuber]

Since this is really a problem of Manager.load_profile maybe the unit test should also be tested directly on that method and not on the load_dbenv_if_not_loaded wrapper function. But the latter should really also be tested. But I would essentially end up duplicating the test. Should I do that nevertheless?

[chrisjsewell]

Suggested change

	if profile is None and self._profile:
	# If a profile is already loaded and no explicit profile is specified, we do nothing
	if profile is None and self._profile:

[chrisjsewell]

I don't think this needs to be removed: it handles for example:

load_profile('x')
load_profile('x')  # no-op

[chrisjsewell]

Should I do that nevertheless?

I think its ok not to

[chrisjsewell]

Perhaps we need to increase the allowed verdi load time from 0.4 to 0.41. For python 3.8, it often can be about this.
I don't think there has been any obvious changes to degrade the load time, in fact I actually changed the check to disallow even loading aiida.storage (formerly aiida.backends)

[sphuber]

Perhaps we need to increase the allowed verdi load time from 0.4 to 0.41.

Yeah, it is indeed a bit annoying, but it would be better still to see if we can get the load time to decrease. I personally at least find it very annoying to have lag in the tab-completion.