RelationalCommand: stricter separation of construction and execution

[divega]

PR #3690 makes the minimal changes required to enable caching and reuse of RelationalCommand: parameter values can now be passed in each individual Execute method call in addition to the constructor.

The change doesn't preclude the existing pattern from being used though, i.e. you create a RelationalCommand passing both metadata and values for the parameters in the constructor, call one of the Execute methods with a null or empty dictionary of parameter values and then discard the RelationalCommand.

Although the option is not exercised in our codebase, it is even possible to use hybrid patterns in which sometimes initial parameter values are used and other times the "fresh" parameter values passed to the various Execute methods are used (but from reading the code I think this won't always work as expected because null parameters passed to the Execute methods are ignored in favor of the corresponding initial values).

@anpete and I had a chat about this and we believe there are potential benefits in terms of code cleanliness and efficiency in forcing all usages of RelationalCommand to separate the construction of the command's parameters from the binding of parameter values at execution time.