Refactor build_model() out of benchmark_model()

[jeremyfowers]

Closes #50 by eliminating the benchmark_model() API
Closes #39 by collecting independent telemetry for builds and benchmarks
Closes #10 by creating the build marker on the proper line of code

Also fixes numerous documentation bugs that were referencing benchmark_model() when they should have been referencing build_model().

Demo

Build Killed

(tkml) jfowers@LAPTOP-5VK03G46:~/turnkeyml/models/transformers$ turnkey bert.py --process-isolation --timeout 10 --rebuild always

Times out during torch export.

in stats: build_status: running. Before, it would have said benchmark_status: running.

Shows in turnkey cache list and gets properly deleted in turnkey cache delete --all. Before, it would have been ignored.

turnkey cache list
Info: Builds available in cache ~/.cache/turnkey:
bert_transformers_bf722986

(tkml) jfowers@LAPTOP-5VK03G46:~/turnkeyml/models/transformers$ turnkey cache delete --all

Info: Deleted build: bert_transformers_bf722986

rebuild=never

After the killed build, run turnkey bert.py --rebuild never

This will throw Build intentionally skipped because rebuild=never as expected.

The stats say: build_status: running, which is the expected value (no mention of benchmark_status). Before, it would have said benchmark_status: failed which would have covered up the fact that the original build was killed.

 

Benchmark killed

Run turnkey bert.py --rebuild-always and then ctrl+c it after the build finishes.

In stats:

    benchmark_status: running
    build_status: successful

Before, it would have just said benchmark_status: running with no mention of build status.

[jeremyfowers]

https://test.pypi.org/project/turnkeyml/0.4.0/ available for testing

[danielholanda]

Tested locally on a few different scenarios.
Working as expected + reducing a ton of code.
Very happy!