Add runtime observation type to LLVM service.

[ChrisCummins]

This adds support to the LLVM environments for optimizing for runtime. To evaluate the runtime, the LLVM-IR module is compiled down to a binary and executed on the host machine multiple times. The wall time of each execution is then returned as a list. This is exposed through a new Runtime observation space.

To compute the program runtimes, use the new Runtime observation space:

>>> import compiler_gym
>>> env = compiler_gym.make("llvm-v0")
>>> env.reset()
>>> env.observation["Runtime"]
[0.123, 0.142, 0.153]

The Runtime observation space returns a list of runtimes in seconds, resulting from running the compiled binary multiple times. Set the number of runtimes to measure using:

# Use 10 runtime observations:
>>> env.send_param("llvm.set_runtimes_per_observation_count", "10")

This supports the cbench-v1 dataset partially and the csmith-v0 dataset fully. To check if a benchmark is compatible with the runtime observation space, use the IsRunnable observation:

>>> env.observation["IsRunnable"]
1

[codecov-commenter]

Codecov Report

Merging #307 (4ae99e3) into development (164cd97) will increase coverage by 0.28%.
The diff coverage is 95.52%.

❗ Current head 4ae99e3 differs from pull request most recent head eb1a4b8. Consider uploading reports for the commit eb1a4b8 to get more accurate results

@@               Coverage Diff               @@
##           development     #307      +/-   ##
===============================================
+ Coverage        85.59%   85.88%   +0.28%     
===============================================
  Files               87       87              
  Lines             4700     4754      +54     
===============================================
+ Hits              4023     4083      +60     
+ Misses             677      671       -6     

Impacted Files	Coverage Δ
compiler_gym/service/proto/__init__.py	100.00% <ø> (ø)
compiler_gym/views/observation_space_spec.py	86.17% <71.42%> (+1.90%)	⬆️
compiler_gym/wrappers/core.py	93.84% <97.05%> (+4.37%)	⬆️
compiler_gym/envs/compiler_env.py	86.70% <100.00%> (+0.25%)	⬆️
compiler_gym/envs/llvm/datasets/__init__.py	100.00% <100.00%> (ø)
compiler_gym/envs/llvm/datasets/cbench.py	78.76% <100.00%> (+2.12%)	⬆️
compiler_gym/envs/llvm/datasets/csmith.py	93.26% <100.00%> (+0.13%)	⬆️
compiler_gym/spaces/scalar.py	96.66% <100.00%> (+0.11%)	⬆️
compiler_gym/spaces/sequence.py	90.00% <100.00%> (+1.53%)	⬆️
... and 3 more

---

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 164cd97...eb1a4b8. Read the comment docs.