Make CAS primops inline
Motivation
When using compare-and-swap to implement read-modify-write (perhaps its main application), it's important to minimize the delay between reading the old value and attempting the CAS—the longer the delay, the higher the risk that another thread will make a change in between, causing CAS to fail and requiring the operation to be repeated.
Proposal
Turn the following into inline primops: casMutVar#, casArray#, casSmallArray#,
casInt64Array#,
casInt32Array#,
casInt16Array#,
casInt8Array#,
casIntArray#, and
casIntArray#.
It looks like the following are already inline, I think?
atomicCasWord64Addr#, atomicCasWord32Addr#, atomicCasWord16Addr#, atomicCasWord8Addr#, atomicCasWordAddr#, and atomicCasAddrAddr#