Ensure cuda functions are correctly jitted

[ianthomas23]

Some of the recently introduced CUDA _append functions in reduction.py were using standard numba jitting not cuda-backed jitting. This PR fixes them.

[codecov[bot]]

Codecov Report

Merging #1153 (6858a3a) into master (b69aaef) will not change coverage. The diff coverage is 100.00%.

@@           Coverage Diff           @@
##           master    #1153   +/-   ##
=======================================
  Coverage   85.37%   85.37%           
=======================================
  Files          34       34           
  Lines        7782     7782           
=======================================
  Hits         6644     6644           
  Misses       1138     1138           

Impacted Files	Coverage Δ
datashader/reductions.py	87.23% <100.00%> (ø)

:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more