Print percent of *tests* where element of a model coverage is involved

[buzden]

Currently, after the #94, count of presence in a model coverage is printed along with coverage status.

Now it looks like this (takes from test here):

PrintCoverage.Y covered partially (100 times)
  - Y1: covered (41 times)
  - Y2: not covered
  - Y3: covered (59 times)

There were 100 tests generated, Y was used in each test once, thus the whole count of usages of the type Y is 100% for the tests, where individual actually generated constructors have 41% and 59% respectively.

But when type is recursive, or just is used several times, one can get percentage >100% (taken from another test here):

PrintCoverage.X covered partially (118 times)
  - X1: covered (54 times)
  - X2: not covered
  - X3: covered (64 times)

PrintCoverage.Y covered partially (100 times)
  - Y1: covered (41 times)
  - Y2: not covered
  - Y3: covered (59 times)

Actually, type X is present in 59% of the tests, along with the Y3 constructor, but each time it is present twice. If we were printing this with a percentage, 118% near to the X type would be a misleading number, where "present in 59% of tests" could be feasible one.

So, it is suggested to add percentage of presence in the tests, but we must be accurate and (unlike the current times coverage) to not to count each model coverage element several times in a single test.

The current design does not support this directly, we don't store this type of data. There are at least two possible solutions to this:

• change the order of ModelCoverage collection so that each test uses registerCoverage independently, and thus is can be changed to calculate percentage of mentions of types and constructors per test; I don't like this solution because of potential high count of calls to relatively expensive registerCoverage function;

• to have two separate Semigroups implemented for the ModelCoverage type: one is for use with the Writer inside appropriate unGen which does not accumulate the usage-per-test metric, and is not exported outside the module; and the other which does accumulate usage-per-test metric and is exported outside; the latter one should be declared in the bottom of the module to not be used accidentally in the Writer context.