`check_for_data_dependent_parallel_bounds` doesn't know about multiple kernel launches

It's perfectly valid for one kernel to compute the launch bounds of the next.

https://github.com/inducer/loopy/blob/9c1e37dc859e1132741025ac05dfa00622b08e0d/loopy/check.py#L486-L509

Start in loopy/check.py at lines 486-509, where check_for_data_dependent_parallel_bounds handles launch-bound validation. Trace how it treats multiple kernel launches and identify the case where one kernel computes the next kernel's launch bounds. Done means valid multi-launch programs are accepted without weakening the existing check.

Written by the indexing model from the issue text.