Improve and distribute epsilon-delta test function across exercises

[sshine]

Before 4c5fd1d, no tests used Expect.nearTo. Since the canonical test data expresses expected values with some unknown imprecision (e.g. 31.69 when the result is 31.68808781), it is up to the language track to make test cases succeed correctly and meaningfully.

Since Standard ML has no standard test framework, we distribute a testlib.sml along with each exercise. The OCaml track uses OUnit2's cmp_float which I've adapted in testlib.sml for the space-age exercise.

This task consists of three parts. Feel free to pick any of these.

Evaluation

• Evaluate if this method is sufficient. See this StackOverflow discussion.

Distribution

• Create a testlib.sml in the root directory that contains Expect.nearTo. This way we can introduce a policy that in order to update testlib.sml, we must also update this reference to it.
• Copy this testlib.sml to all existing exercises; even though they don't depend on epsilon-delta tests, it is easier to have only one version of testlib.sml.

Assertion

• Create a CI script that asserts that all testlib.sml in the repository have the same checksum.

[guygastineau]

Oh, hah, I found this discrepancy in the tests from my other work on #206 and #205. I guess I have done everything except for the final step. Maybe I should add a script and CI action to verify the checksums of all testlib.sml files to #206 in order to satisfy this issue entirely.

[guygastineau]

@sshine Do you think our current implementation of nearTo is good enough, or should we use the full example you posted in the linked post? I don't really care enough to scrutinize their differences too much. FWIW, nearTo as it is in the testlib.sml for space age is immediately obvious.