Remove experimental Throttle api.

[jpinner-lyft]

This feature was never completed and has several issues as currently written:

• it does not distinguish read vs write capacity (Throttles do not distinguish between read and write operations #86)
• records are a mix of capacity units and consumed capacity maps
• there is no distinction between table capacity and index capacity (Rate limited scans and Throttle do not work with GSIs #239)
• query and scan operations do not set return_consumed_capacity on the low-level api calls

[garrettheel]

LGTM