Revisit lock-free ThreadLocalStorage storage management

[jonwis]

https://github.com/microsoft/wil/blob/master/include/wil/result.h#L403-L430

Internal analysis suggests that this sequence is not entirely thread safe. Some options include:

• Use an srwlock-per-bucket (costs a function call)
• Figure out the right sequence of acquire/release semantics for reading the hashtable bucket entries' starting pointer
• Spin gently (using the platform's "yield" instructions as necessary)
• Deeper analysis to prove safety with annotations and documentation

Thanks to @sudhakar-vp-zz and @cpkleynhans for the analysis! (This is http://task.ms/58950594 for future explorers.)

[jonwis]

See potential approach for "lock per bucket" here - https://gist.github.com/jonwis/f50d8f3d28a85a074cb74aa13d4c774c ... the additional AcquireSrwLockShared call to read the bucket field is "cheapish" and the "Exclusive" happens once per thread on its bucket.