This PR experiments with adding a new executor which can execute benchmark variants according to "programs".

A program is a nestable sequence of named "code snippets"/units which are used to compose the benchmarking script.

Benefits:

• Full flexibility with how the script is executed.
• Potential to use different strategies (e.g. instead of doing a for loop over the revs, print out the call N times)
• Potential to add more samplers (not sure what those would be outside of memory/time though)

The default PHPBench executor can be reproduced with:

[
            'before_methods',
            'warmup',
            'hrtime_sampler',
            [
                    'hrtime_sampler',
                    'call_subject',
            ],
            'after_methods',
            'memory_sampler"
]

POC TODO:

• Create a POC lexer/parser for programs
• Generate scripts/programs from units
• Create new PHP process factory / script executor
• Map script results to classes automatically (support collecting as many different metrics with the same executor)
• Create the necessary units (hrtime, memory sampler, etc) .
• Support HRTime

TODO:

• Is there a viable alternative to the DSL? YES: Replaced with a simple list/object representation
• Test everything, ensure programs can be constructed safely in any order, enforce cardinality of units (i.e. max 1 of each sampler type)
• ...

IDEAS:

• Perhaps opcache can be supported by keeping the benchmarking script(s) for the duration of the run?
• Support ephermal (file-less) benchmarks, pass the code directly to the PHP process?