Improve clean up on user disconnection

What's Changed

• Update to latest builder by @bretambrose in #209
• Fixed endpoint not being read in arguments by @sbSteveK in #211
• No resubscribe by @bretambrose in #215
• Clean up on user disconnect by @sbSteveK in #216

New Contributors

• @sbSteveK made their first contribution in #211

Full Changelog: v0.7.10...v0.7.11

Remove using of try_lock

• Try lock is not available in Windows Server 2008. Remove it from our debug build

Empty payload fixes

• Fix a crash on Windows when the will payload was empty (precondition is turned into a fatal assert on Windows CD builds still)
• Fix a decode error when receiving a message with an empty payload

Stability fix

• Fixes a crash due to a race condition in the mqtt connection logic

Allow for null on-message callbacks in subscribe

• Fixes a crash when the on-message callback is null for a subscription
• Updates the public API documentation with some missing parameters and annotations as to what parameters can be null

Data race fixes

• Fixes several data races related to the outstanding requests table and a request's timeout task.

Fix non-threadsafe hash-table accesses (take 2)

v0.7.5

fix test function (#190)

Fix non-threadsafe hash-table accesses

Fix unsafe hash-table accesses (#189)

Fact: It's not safe to mutate aws_hash_table while holding an aws_hash_element.
Fact: `outstanding_requests_table` is accessed by multiple threads. We always hold a lock while accessing it.

Bug: There were several places where we release the lock, but continue to hold an aws_hash_element we got from a find() call. Once the lock is released another thread can mutate the `outstanding_requests_table`, leading to undefined behavior when the aws_hash_element is accessed again.

Direct mqtt through http proxy

• Adds support for establishing direct mqtt connections through an http proxy

publish payload

• We now keep a deep copy of the publish payload. User don't need to keep the payload alive for the whole process.