Recognizers should register their names to allow dynamic pipeline creation.

[MikeHopcroft]

This enables a couple of scenarios

1. Test files can include a pipeline specifier that lists the recognizers required for the test. This allows testing of individual recognizers and combinations of recognizers.
2. The CompositeRecognizer can produce more specific logging information by naming the recognizer at each stage.

[MikeHopcroft]

With this feature, recognizers could also add their names as prefixes to logging messages. Right now we see

14 items contributed 143 aliases.
5 items contributed 22 aliases.
16 items contributed 31 aliases.
60 items contributed 208 aliases.

instead of

EntityRecognizer: 14 items contributed 143 aliases.
AttributeRecognizer: 5 items contributed 22 aliases.
QuantityRecognizer: 16 items contributed 31 aliases.
IntentRecognizer: 60 items contributed 208 aliases.

[MikeHopcroft]

CompositeRecognizer should also use the names for logging.

[MikeHopcroft]

We no longer have a recognizer pipeline, but the comments, above, may still apply to domains.