[AArch64] match fcvtl2 with bitcasted extract

This should eliminate a regression seen in D63815.

If we are FP extending the high half extract of a vector,
we should be able to peek through a bitcast sitting
between the extract and extend.

This replaces tablegen patterns with a more general
DAG to DAG override, so we can handle any casted type.

Differential Revision: https://reviews.llvm.org/D71515