Fix CUDNN-optimized activation broadcasts

[maleadt]

Fixes:

julia> x = CUDA.ones(1); tanh.(x); x
1-element CuArray{Float32,1}:
 0.7615942

julia> x = CUDA.ones(1); tan.(x); x
1-element CuArray{Float32,1}:
 1.0

Introduced by #321, discovered while debugging FluxML/Flux.jl#1367

[codecov]

Codecov Report

Merging #515 into master will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #515      +/-   ##
==========================================
+ Coverage   80.52%   80.54%   +0.02%     
==========================================
  Files         166      166              
  Lines        8884     8884              
==========================================
+ Hits         7154     7156       +2     
+ Misses       1730     1728       -2     

Impacted Files	Coverage Δ
lib/cudnn/activation.jl	100.00% <ø> (ø)
lib/cudnn/tensor.jl	59.52% <ø> (ø)
test/cudnn.jl	26.66% <ø> (ø)
lib/cudnn/nnlib.jl	70.00% <100.00%> (+1.81%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c70bb39...beffb7d. Read the comment docs.