Datashader internals to-do list

[jbednar]

Immediate fixes needed:

• 'shade' function doesn't correctly implement 'span' #368: Fix span argument (PR Fix handling of the span argument to tf.shade #680)
• Additionally re-run (some part of) the test suite without numba #646: Make part of test suite run without Numba to catch array bounds and similar errors (PR Fix index-out-of-bounds line glyph crash and run select tests with JIT disabled #683)
• Datashader crashes python with "noisy" timeseries data. #570: Datashader crashes for long lines (floating-point issue) (PR Fix index-out-of-bounds line glyph crash and run select tests with JIT disabled #683)

Relatively urgent fixes:

• KeyError 'get' #641: Fix dask get issue? (PR Migrate to dask 1.0 public API and remove legacy workarounds #681)
• Memory errors on distributed dask cluster #668: Memory errors on distributed dask cluster
• Improve colors for small numbers of discrete values #357: Improve colors for small numbers of discrete values

Less-urgent fixes:

• canvas.raster mean aggregation on integer arrays causes artifacts #655: Fix non-count aggregation on integer arrays
• Handle categorical columns correctly on dask dataframes #653: Fix categorical columns on dask dataframes
• Dask dataframe chunksize/partioning affects line aggregation #464: Line aggregations don't respect Dask chunking (urgent but only easily addressable with ragged-array support)

Less-urgent new features:

• Generalizing support for categorical data #140: Generalize count_cat to aggregate any field categorically, not just counts
• Add cyclic aggregator #630: Cyclic aggregators. Implemented in category_modulo and category_binning #927.
• Improving Datashader's API #490: Improve datashader API (bigger task)
• datashader memory consumption on very large dataset #638, Largest Canvas size allowed? #409, Is it possible to improve the performance of line()? #456, Using distributed scheduler with persist leads to slower loading times, despite dataset fitting into RAM with threaded scheduler #337, Performance degradation of repeated aggregations with distributed scheduler and client.persist() on a single machine #332: Benchmark (capturing with asv?) and improve performance and memory usage