Estimate improvement

[bazhenov]

This PR adds following changes:

• there was asymmetry in benchmark haystack generation which favors one of the functions
• benchmarks seeds are now synchronized after estimation process to guarantee equivalent PRNG state in both functions
• time slice changed to 10ms
• new iteration count estimation process adaptively changes the number of iterations to guarantee 10ms per sample.
• warmup iterations are performed before each sample (1/10th of sample iterations). Can be disabled with --warmup-enabled=false
• -t option now take into consideration only tested function execution time. No additional timer overhead is required.
• timer precision added. If time difference between functions is less than absolute precision timer test results considered not significant