Redesign hybrid accuracy test

[XingerTang]

The hybrid mode is designed to produce results similar to multi-locus peeling, but faster.
While the current accuracy tests for hybrid mode have the following problem:

• The time measured is only the single-locus peeling part of the hybrid peeling, and does not count for the multi-locus peeling part of the hybrid peeling
• While we already measured much less time than we should, the runtime for hybrid mode is still quite long, or sometimes even longer than multi-locus peeling
• The problem is that in a real scenario, we may reduce the peeling cycle for the two stages of the hybrid mode for efficiency, while in the current accuracy tests, it always uses the default number of cycles

The tests should be redesigned in a way:

• The runtime of the multi-locus stage is also measured
• Fewer cycles for both stages to mimic a realistic use case

[XingerTang]

I come back to this issue, and now I think it is not the problem of the peeling cycle but how we set the inputs.

For the accuracy tests:

The current hybrid peeling is not doing what it should do; currently, it is running

• multi-locus peeling on full input genotype data/sequence data and then subset the generated segregation probability output file to get the segregation probability on only the specified map sites
• run the single ("hybrid") locus peeling based on the subset segregation prob, and only benchmark this part

But what it actually should do is:

• subset the genotype data/sequence data and then run the multi-locus peeling and get the corresponding segregation probability output
• run the single-locus peeling with the segregation probability output from step 1 as well as the full genotype data/sequence data, and benchmark both steps for the runtime

[gregorgorjanc]

Good catch @XingerTang!