v0.8.3

Performance:

• Up to ~20% faster pool initialisation
• New worker signalling system reduces task submission overhead
• Reduced some atomic orderings backed by formal verification (Still valid on weak memory models, eg: ARM)
• Inlined user facing TaskFuture methods (wait, is_complete, wait_timeout)
• Cold-path separation of retired list cleanup logic to minimise worker loop overhead

Concurrency & Memory:

• Resolved MIRI aliasing/tree-borrows warnings during pointer retagging (Only affected internal tests)
• Prevented potential duration underflow panics in wait_timeout using saturating_sub

Benchmarking:

• Split benchmark file into dedicated targets for accurate performance isolation

• Readme updated