…ment() (AztecProtocol/aztec-packages#6523)

AztecProtocol/aztec-packages#6405 added a very
complicated call (`get_npk_m_hash`) to `destroy_note`, not realizing
that `destroy_note` is called in a loop by `decrement`. The loop
unrolling caused multiple calls to this function, even though they all
had the same arguments, ultimately resulting in humongous RAM usage
during compilation (over 40GB just for this contract) and very
compilation times (from 30s to multiple minutes).

This PR fixes this simply inlining the code for `destroy_note` and
fetching the key once at the beginning. It does worry me slightly
however that such a large performance hit was not noticed - this likely
affected test times as well.