Dynamic timeouts for engine based fuzzers, also analyze_task timeout should be higher (not 25 sec)

[inferno-chromium]

Three bugs here:

1. Too low timeout in google instance template. Now increased to 90 secs.
2. Fix timeout argument during reproduction (upload testcases are not impacted as they dont have timeout arg at all).
3. Use a fuzzing strategy for -timeout argument so that we try higher values than 25 sec for some runs.

[inferno-chromium]

1. and 2) are fixed.

[nwellnhof]

AFAICS, reproduction now uses TEST_TIMEOUT-5 which results in 60 seconds on OSS-Fuzz. If I specify a custom timeout in the .options file, for example 20 seconds, then a test case taking 40 seconds will initially result in a crash but the crash will be deemed non-reproducible. This also means that fixing the timeout won't be detected.

I think it would be better if reproduction used the timeout from the .options file, multiplied with a factor like 90%.