refactor random crate search, the random view size is configurable and we don't retry any more

[syphar]

The test-database has not very many test-crates in the id-space which have > 100 github stars, so the random-crate-search for I'm feeling lucky fails in the tests.

This refactors the random crate search so the major factor for its success can be configured.

[jyn514]

Thanks for the PR. Unfortunately, the "feeling lucky" search still occasionally fails.

> i=0; while cargo test lucky; do i=$((i+1)); done
...
thread 'web::releases::tests::im_feeling_lucky_with_stars' panicked at '/releases/search?query=&i-am-feeling-lucky=1: expected redirect to /some_random_crate/1.0.0/some_random_crate/, got 404 Not Found', src/test/mod.rs:80:13
> echo $i
62

But it seems a lot better than before, so I'm ok merging this as a temporary measure.

[syphar]

@jyn514 if nothing serious is blocked because this, give me a little more time today in the evening. I'll figure something out.

Flaky tests annoy the hell out of me when I stumble on them, I don't want to be the reason for them :D

[syphar]

re-adding the retries would help for the tests, but that doesn't feel right.

It's tricky with that kind of test-setup, the frameworks I'm used to currently isolate the data for single test-cases.

[syphar]

After discussion in the channel we came onto the issue with this test, driven by the test-setup in 44f98be and me relying on a relation between data and the primary key sequence.

[jyn514]

I ran this 145 times in a row without issues, nice job :)

[syphar]

@jyn514 thanks for reviewing and merging!

And sorry that it broke :)