Fix race conditions in censor's querywriter

[Lagovas]

It is the second approach to fix flaky tests in acra-censor package by fixing race conditions in querywriter. What was done:

• all operations over writer.queries wrapped with RW mutex to avoid race conditions and make it more predictable in tests
• added -race flag for unit tests to detect race conditions. we used to old version of boltdb that are not maintained more. And it has race condition issues detected with race detector. So I updated boltdb dependency to new one that hasn't such issues (etcd-io/bbolt#187).
• split unit-tests into two parts of running. one tests with previous versions with golang as it was before. And one more with latest version that run tests with -race flag. Race detector increase time of tests and actually we don't need to run it trice with this flag. We can use latest version that should have the newest updates in go compiler.
• removed -v flag from go test command because of it generates 400k+ log lines that we don't use actually. It's enough to use default mode where will be logged failed test with name and line of error. And we would need to download 400k log lines to grep failed test name.
• removed extra run of unit tests with GODEBUG=tls13=1 because of it doesn't work anymore since go1.14 where tls1.3 support turned on by default.

Checklist

• [x] Change is covered by automated tests
• [x] The coding guidelines are followed
• [x] Public API has proper documentation in the Acra documentation site or has PR on documentation repository with new changes
• [x] CHANGELOG.md is updated (in case of notable or breaking changes)
• [x] CHANGELOG_DEV.md is updated
• [x] Benchmark results are attached (if applicable)
• x are up-to-date (in case of API changes)

[iamnotacake]

If it's possible, in future we could add files like the CHANGELOG_DEV.md and CHANGELOG.md to CI blacklist to avoid triggering it. Or simply add [skip ci] to such "text only" commits' titles to avoid wasting resources.