reorganize time/ to benchmark/ and set up CI

[johnnychen94]

• test related codes are removed; code logic should be tested in test stage instead of benchmark stage
• remove BenchmarkTools dependency from Project.toml; this package should not be a dependency of MIRT
• use @benchmarkable instead of @btime as the former can be used by BenchmarkPkg and BenchmarkCI
• Set up a Benchmark CI that get triggered when you label the PR with run benchmark

The benchmark CI could error in this PR if gets triggered because currently benchmark/ doesn't exist in master branch

[codecov]

Codecov Report

Merging #50 into master will decrease coverage by 0.04%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master      #50      +/-   ##
==========================================
- Coverage   98.86%   98.81%   -0.05%     
==========================================
  Files          48       48              
  Lines        2107     2107              
==========================================
- Hits         2083     2082       -1     
- Misses         24       25       +1     

Impacted Files	Coverage Δ
src/mri/mri_objects.jl	99.38% <0.00%> (-0.62%)	⬇️
src/plot/jim.jl	100.00% <0.00%> (ø)
src/fbp/sino_geom.jl	100.00% <0.00%> (ø)
src/fbp/image_geom.jl	100.00% <0.00%> (ø)
src/mri/mri_trajectory.jl	100.00% <0.00%> (ø)
src/utility/downsample.jl	100.00% <0.00%> (ø)

---

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 56ca418...8bc50d0. Read the comment docs.

[JeffFessler]

@johnnychen94, the julia code in the time/ directory is some simple timing tests to compare different versions of certain operations like downsampling. After I run those tests during development and determine which version is faster, then I just export the faster version in the package. I like to leave the timing code in the package as a reminder to myself for later how I chose between different options, and I also thought maybe someone else would want to test to see if the version I chose was faster on their platform and if not they could report an issue if a different version is faster.
At first I thought this PR would cause a benchmark run in the cloud every time we do a PR, which would waste energy. Now I understand that it will be invoked only with run benchmark as a label, so that is reasonable.
However, I do want to keep the files I had in the time/ directory so I will restore those after merging.