Rearrange tests for XOnlyPublicKey::from_slice

[bigspider]

Adds two test cases for parsing an invalid key (despite it being 32-byte long).

Perhaps the test for ZERO is redundant, as 7 also has no square root, but it felt natural to add it anyway.

[apoelstra]

The CI failures are because when you compile with cfg secp256k1_fuzz, these do parse correctly, since with fuzzer mode on we don't actually parse public keys; we parse a key-like object. (Where "key-like" means "parses super fast and behaves enough like a key that it doesn't crash downstream stuff.)

I think the best solution is to add these tests as a new unit test with #[cfg(not(secp256k1_fuzz))] marked.

[bigspider]

Actually, I just saw that there is a test_pubkey_from_bad_slice a few tests below, so I think my PR is pointless.
It might be worth moving them next to each other, and perhaps delete a few redundant test cases.

[bigspider]

I did the slight refactor (without deleting any test) - feel free to merge or close as you please.

[apoelstra]

ACK a4cb345; successfully ran local tests; yeah, sure. I think this makes the tests a bit easier to follow