Thread safety enhancements

[Stephen-Bao]

This is the equivalent PR as #104, but pointing to the master branch.

To track conversation history, check #104.

Issue #, if available: #61

Description of changes: Added thread-safety enhancements, unit tests, and benchmarking code.
The benchmarking code measures the performance of the library under 3 diffrent internal synchronization situations:

1. no internal sync for flush()
2. using ReentrantReadWriteLock to make flush excute exclusively
3. using Stamped to make flush excute exclusively

The main purposes of the benchmarking are:

1. measure the extra overhead introduced by adding an extra lock on single-threaded users
2. gather information to decide whether to provide a separate synced flush() while preserving the old one
3. compare the performance of ReentrantReadWriteLock & StampedLock

Here are the benchmarking results:

1. Single-threaded Use Cases

This proves that adding either lock would not cause an significant extra delay to single-threaded users.

2. Single Method Performance in Async Context
   a. each thread publishes 1000 metrics

b. each thread flushes 100 times

These tests indicates that internal sync with a lock increases average ops time on single method, but overall ops time is still fast enough.

3. Mixed Methods Performance in Async Context
   Each thread executing mixed metheds 100 times:

This illustrates the overall performance under different lock situations, especially the performance difference between the two locks.

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.