Add support for Long and Float numeric dimensions (without indexing)

[jon-wei]

This PR adds support for Long and Float numeric dimensions.

It depends on the following two PRs:

• Allow IncrementalIndex to store Long/Float dimensions #2263
• Add DimensionSchema class for specifying dimension properties druid-io/druid-api#74

Implementation notes:

• Dictionary encoding is not used for Long and Float dimensions. No index structures are currently created for numeric dimensions. Numeric dims will use the existing long/float column formats on disk.
• The DimensionSelector interface now has additional functions for retrieving rows from numeric dimension columns without dictionary encoding.
• A new DimensionSelector, UnencodedDimensionSelector, has been added for use with numeric dims.
• As there are no bitmap indexes on numeric dims, a full scan will be performed on numeric dims for filtering logic that uses filter.getBitmapIndex()
• To support the requirement above, ColumnSelectorBitmapIndexSelector now supports bitmap generation from the result of a column scan
• SelectorFilter and BoundFilter can do direct numeric comparisons on Long/Float columns, no string conversion needed
• Long and Float values will be converted to Strings for any filters or extraction functions that require String inputs.
• null values for Long and Float dims will be converted to 0L or 0.0F, respectively.

Unit test notes:

• The druid.sample.tsv file has been modified to include Long and Float dimension columns (market_long, market_float, quality_long, quality_float). The original druid.sample.tsv file has been kept in the repo.
• The numeric values for these new dimensions have a 1-to-1 mapping with the values in 'market' and 'quality'
• For larger test suites like GroupByQueryRunnerTest and TopN, I have parameterized the use of the "quality" and "market" dimensions; in different test iterations, the tests will use the long or float versions of these dimension columns.
• Smaller test suites have separate functions for testing numeric dimensions.

[fjy]

@jon-wei this is failing UT

[jon-wei]

@fjy This PR won't be able to pass UT as-is, it depends on a change to DimensionsSpec in druid-api:
druid-io/druid-api#74

I'm opening it now to get the review process started

[jon-wei]

Based on a discussion with @gianm this afternoon, I'm looking into abstracting out a generic/extensible "dimension handling" interface across the IncrementalIndex/IndexMerger/querying paths.

Will update this PR as I make progress.