Bugfix: Clean up ThreadContext pointers when Connection is destroyed #106

Currently ThreadContext Connection* pointers are not removed up when a
connection is destroyed. This is only a problem if a Connection instance is
destroyed and new Connection is allocated at the same address, because the code
assumes pointers uniquely identify connections. This causes a bug in a bitcoin
IPC test which creates multiple connections in a loop, described in
https://github.com/bitcoin/bitcoin/pull/30509#issuecomment-2276739868, where
connections are created and destroyed in a loop, and depending on how the heap
allocator behaves, a new Connection could have the same address as a previously
destroyed connection, and the code tries to use a thread reference associated
with the previous connection when making a new call, and there is a segfault
because the thread no longer exists.

Fix this problem by adding Connection cleanup callbacks to remove Connection*
pointers from the ThreadContext struct if the connection is destroyed before
the thread is.