JIT: fix bad assumption in non-funclet EH models

JIT: fix bad assumption in non-funclet EH models

The newly added finally optimizations mistakenly assumed that in
non-fuclet EH models the GT_END_LFIN in the continuation block would
be the last statement. The test case added below provides an example
where this is not so.

Relax the assumption and instead search the continuation for the
GT_END_LFIN. Assert that there is exactly one.

Commit migrated from https://github.com/dotnet/coreclr/commit/0df4892d1d7e04fc817beb7f10be4faae0dfaf82