JBrainy now supports bytecode benchmark generation, and
benchmark generated based on experimentally derived Markov chains.

Data is checked and parsed before doing normalization.

This data is from using markov chains for generating benchmarks

This data is from using markov chains for generating benchmarks

There was a null pointer exception when trying to generate
a benchmark from a record where the method selection string
was set to NULL.

Required some refactorings.

Created a type for interface type: InterfaceName
Created a type for method selection type : MethodSelectionType

Refactored benchmark generation for automatically loading markov
chains if needed, only the method selection type is needed to
create a benchmark generator.

The column "Param: MethodSelectionStrategy" is parsed and a proper JMH record is created.

These markov chains are generated with Pomegranate from traces on
the fop, lusearch, avrora benchmarks

Since some data from this class was used, I moved the method in this
class

This test checks that genBenchmark does generate benchmarks which are
different when the markov chain is loaded

Tested loading markov chains with standard java method names
and using them as a method selection strategy

There were quite a few things wrong with the build script (remnants of
build scripts for older versions of gradle)

For each new benchmark, we need to reset the markov
selection strategy, this is implemented in the reset method.

The setup is a bit convoluted, to allow for loading the
markov chains from files so that you can use them for benchmark
generation:

There is a static map indicating the markov chain for each interface
type (List, Map, or Set), with another static map to fetch the
markov chain if it's already loaded when generating the benchmark

Trying to make the `jmhJar` task work.

A benchmark is generated from a markov chain and we test that
the benchmark's trace looks exactly like what we would expect.

A markov selection strategy can now be used to run a benchmark
where the markov chain is serialized in JSON

The constructor now takes the MethodSelectionStrategy as
a parameter. Code in tests and benchmarks is updated
to adapt to the refactoring.

The label loading from JSON works, completing the implementation.

The part of the test regarding transition probabilities works

Loading the initial state probabilities works.

Markov chain based benchmarks are guaranteed to be correct since
each new method selected is checked to be a valid one.

The data_seed is now used, instead of a magic constant.

Added test to check Set benchmark generation also works as expected

Added test to make sure the BenchmarkGenerator for Map benchmarks also
works as expected.

The MethodProbabilityTable does not do generation on it's own anymore
this functionality is delegated to the MethodSelectionStrategy.

The table is just used as table of probabilities.

(Eligibility is still to be checked, though)

Currently fails (the seed is ignored)

With tests

Markov chain now has an internal state and allows for sampling one sample at a time.

The sample() method allows to sample from a markov chain, with tests