fix: randomID returns shorter IDs with same amount of entropy

[peterldowns]

As reported in #16, the previous implementation of randomID used a hashing construction (a) inappropriately, (b) incorrectly. As a result, the IDs that were being generated had a random prefix and then a consistent suffix, like this:

3aa05350d41d8cd98f00b204e9800998ecf8427e
f1c841c1d41d8cd98f00b204e9800998ecf8427e
e5dde3f7d41d8cd98f00b204e9800998ecf8427e

This PR fixes the function to essentially just return the randomly generated prefixes, and omit the suffix entirely. The fixed-suffix was always a mistake and never intended, I just wasn't paying close attention beyond "are the instance database names colliding".

I tested this change by running some tests that used pgtestdb and intentionally failed, then checking the connection strings that were printed as part of the failing test's logs. Here's an example:

testdbconf: postgres://pgtdbuser:pgtdbpass@localhost:5433/testdb_tpl_c2b33a22a68750f84b5af76e9774be0e_inst_a4a3ce6c?sslmode=disabl

The key part is inst_a4a3ce6c, which is now (a) shorter than before, (b) just as random.