* Each test that uses a topic now creates its own.

Topic names use test name and unix time, so it is easy to
find and inspect it when the test fails. This also removes
some offsets that worked around other tests having polluted
the common topic.

* Do not depend on values being shared between assert closures.

* Flag to return messages received so far when maxWait is reached.

The current behavior, to panic on deadline is problematic, since
there are a number of scenarios where you simply do not know how
many messages there are to consume or when they are produced. For
example, a load test may produce messages in bursts with jitter
for a burn-in test, to try to provoke i data races in the test
subject.

This commit introduces a flag ExpectTimeout which will return message
received so far.

The commit also adds a test to verify that we panic if Close() is
called while we are consuming. Given the nature of the problem
xk6-kafka is trying to solve, this sounds like an errar condition.

* Include expectTimeout in docs.

Go durations are configured in nanoseconds, however exported timeout metrics have to be in milliseconds. Conversion was missing before for the writer timeouts, leading to timeouts being displayed as 1000000× too long.

* Add extra constants to ensure compatibility with docs
* Parse offset if provided, otherwise use constants
* Update README
* Wrap error and decrease log level to warn
* Clarify
* Update deps
* Fix dep error
* Fix indentation
* Update direct deps