Redesign dictionary id use, add columnFromValues.

[jheer]

This PR redesigns how dictionary ids are used.

• Types: The dictionary() type constructor will set the id to a default value of -1 if not specified.
• Decoding: Dictionary ids are decoded and written to dictionary types by tableFromIPC.
• Encoding: The schema is analyzed and (potentially new) consecutive ids are assigned to each dictionary by tableToIPC. This effective serialization of tables that may have been built with conflicting dictionary ids.
• Building: Builders will treat a non-negative dictionary id as a signal for potential reuse, allowing the same backing dictionary values to be used across multiple columns. In this case, clients must take care not to provide overlapping ids for semantically separate dictionaries.
• Drop dictionaryTypes map from schema objects, as it is no longer needed. This also simplifies some of the decoding code.

In addition, this PR adds a new builder method that allows values to be provided by a visitor function:

• Add columnFromValues method. Instead of providing a materialized array, callers can provide a total length and a visit (or "scan") function that invokes a callback with each successive value.