Auto merge of #1320 - Xanewok:translate-tests, r=Xanewok

Translate remaining tests

Continuation of #1244.

Here are some anecdotal benchmarks (used `time` under Xubuntu 18.04, AMD Ryzen 2600).

* master branch

`$ time cargo test --release client_` 5.5 real (2.5 sys)
`$ time cargo test --release test_ `8.65 real (3.2 sys)

* translate-tests branch

`$ time cargo test --release client_` 5.7 real (6.3 sys)
`$ time cargo test --release test_` 1.3 real (0.8 sys)

Unfortunately it seems that two tests post-translation fail very often (which I disabled for now):
- client_hover_after_src_line_change (sometimes has to fall back to Racer, which is bad)
- client_find_all_refs_test (sometimes doesn't pick up reference from `#[cfg(test)]` code - maybe analysis chokes when coalescing defs from the 2 crates with same origin but different cfgs?)

This is alarming since the new approach should isolate the behaviour better (we start a new process in a scratchpad directory), which means that either I'm not seeing what I'm doing wrong with the new approach or it uncovered some bugs with the rls-analysis (seems like an obvious culprit for both of these failures).

r? @alexheretic

Author: bors