Optimize `reduce`

[dmehra]

Add optimization for read opentsdb ... | reduce, including:

• reduce count(), count_unique()
• reduce avg(), min(), max(), sum() - some adapters may optimize more exotic reducers too
• reduce by
• reduce by
• reduce -every [-on]

Or whatever OpenTSDB will permit us to optimize in reduce. Looks like their 'aggregators' should handle our standard reducers, the by-grouping comes native (and would need to be disabled if we're not doing reduce by?), and reduce -every is their 'downsampler'.

[demmer]

reduce count(), count_unique()

These feel like strange optimizations to actually care about in a timeseries database.

[dmehra]

The count thing is handy for debugging ingest path. On count_unique agreed, not useful.