Dataset-scope tailoring of EnsureCommandParameterization validation

[mih]

A dataset often provides a critical piece of information to be able to validate properly and in-depth. For example, the name of an existing remote can only be validated once a dataset is determined in which this remote exists.

Constraints have for_dataset() that is meant to be used for this purpose.

EnsureCommandParameterization would need to call this method for any constraint in order to obtain the most adequate variant.

It could make sense to have the validation order be significant, i.e. have no Dataset identified, until a dataset= argument is processed, only only from there on call for_dataset() on the subsequent constraints.

It may also be necessary to support an alternative name declaration for a "dataset" command, in case one and the same command has two arguments of this type.

For that reason, it may also be beneficial to ignore the name of an argument and go by constraint-type (i.e. EnsureDataset).

Combined with validation order rules (a class author sets the order of arguments in _validator_) it may be sensible to start tailoring constraints when a first dataset-type (not name) argument is found, tailor all constraints from there on, and switch the dataset for tailoring, whenever another dataset-type argument is found.

[mih]

With #234 in place, the situation presents itself to me as follows:

A simple implicit or explicit order-based tailoring is not sufficient. Any particular command may take more than one Dataset-type parameter (e.g. a context dataset (something to work in), plus a dataset target (something to create/delete/modify). Any other command parameter might be related to the former or the latter. Imposing particular naming patterns on all DataLad commands is needlessly complex and likely to cause incomprehensible behavior.

Instead, I propose this:

EnsureCommandParameterization receives a new constructor argument tailor_for_datasets. The value for this parameter is a mapping of parameter names to parameter names. The keys are Dataset-type parameters that provide a context for tailoring other constraints to (see Constraint.for_dataset()), and the values are the parameters that shall have their constraints be tailored for the respective dataset context.

Based on this input, EnsureCommandParameterization can determine which constraints to evaluate first, and which constraints to tailor to what context.

[mih]

I decided to not address this in #234 directly, but have that be merged first.