Run tests in parallel

[Gerold103]

For that it is required to use zero port to bind, and deal with filtering out all the URIs from result files.

[Gerold103]

It is not as trivial as I thought - URIs are fixed in config. Possible solution - start tarantools on zero port, collect real ports, fill them into a config, use that config.

[Totktonada]

use_unix_sockets and use_unix_sockets_iproto options did help me for graphql (if you are about flaky fails). There is also an issue in test-run to fix that: tarantool/test-run#141 (but nobody will do it, I think).

[Gerold103]

@Totktonada - not about flakiness. It is about a necessity to write all the URIs into configuration tables/files before Tarantool start. If do so, 1) parallel tests would collide on the same ports, 2) ports could be occupied already by not Tarantool. Probably unix sockets could help, if they are stored separately for each test in var/ dir. Are they? Then I could hardcode them into configs, and at the same time do not bump into collisions of different tests.

[Gerold103]

This #371 will solve the problem. In the current diff-tests it is not realistic. Will need to rewrite them fully anyway for that.