An initial attempt was made here: #137316 - performing a subvector partial reload of a larger vector spill store - but that could lead to STLF issues.
What we need to do is perform the extract+store at the spill stage so we're only spilling/reloading the required subvector.
