update corpus tests in 3.3.x

[khieta]

Issue #, if available:

Description of changes:

Same as #7, but for the 3.3.x branch. Also includes a freshly-generated corpus, which is distinct from the one in that PR.

[cdisselkoen]

Should we be concerned that the new corpus tests are 85% smaller than the ones checked in?

[khieta]

Hm, I'll gather some stats about the generated tests & post results here before I merge.

[khieta]

Your question prompted me to write a script to check this in the future, which I put up in a separate PR to main (#9). According to that script, here is the diff between the old and new corpus:

	Original	New	Difference
Total test cases	6065	5324	-741
Unique policies	3486 ( 57.5% )	2826 ( 53.1% )	-660
Unique schemas	2298 ( 37.9% )	2163 ( 40.6% )	-135
Unique entity stores	2826 ( 46.6% )	2455 ( 46.1% )	-371
Entities per entity store	mean: 1 / median: 1 / p90: 4	mean: 2 / median: 2 / p90: 4	n/a
Schema entity types	mean: 1 / median: 1 / p90: 2	mean: 1 / median: 1 / p90: 2	n/a
Schema actions	mean: 1 / median: 1 / p90: 2	mean: 1 / median: 1 / p90: 1	n/a
Trivial policies	2191 ( 36.1% )	2259 ( 42.4% )	68
Pure RBAC (non-trivial) policies	1148 ( 18.9% )	910 ( 17.1% )	-238
Pure ABAC (non-trivial) policies	156 ( 2.6% )	111 ( 2.1% )	-45

So the corpus has indeed become smaller compared to the old data (which was from several months ago). This could be due to performance regressions in Cedar leading to fewer tests being executed, or just changes to the fuzzing infrastructure in cedar-spec. Hopefully using the new script in #9 will allow us to track this better.