Datashader internals to-do list

[jbednar]

Immediate fixes needed:

• [x] #368: Fix span argument (PR #680)
• [x] #646: Make part of test suite run without Numba to catch array bounds and similar errors (PR #683)
• [x] #570: Datashader crashes for long lines (floating-point issue) (PR #683)

Relatively urgent fixes:

• [x] #641: Fix dask get issue? (PR #681)
• [x] #668: Memory errors on distributed dask cluster
• [x] #357: Improve colors for small numbers of discrete values

Less-urgent fixes:

• [ ] #655: Fix non-count aggregation on integer arrays
• [x] #653: Fix categorical columns on dask dataframes
• [ ] #464: Line aggregations don't respect Dask chunking (urgent but only easily addressable with ragged-array support)

Less-urgent new features:

• [x] #140: Generalize count_cat to aggregate any field categorically, not just counts
• [x] #630: Cyclic aggregators. Implemented in #927.
• [ ] #490: Improve datashader API (bigger task)
• [ ] #638, #409, #456, #337, #332: Benchmark (capturing with asv?) and improve performance and memory usage