Documentation of `cupyx.lapack.posv()` is inaccurate

Discovered during another internal discussion 😅 This line in the docstring:
https://github.com/cupy/cupy/blob/89430ec6c80373177bb5f8a58fa55f29bc08910d/cupyx/lapack.py#L274
is only applicable to [`cusolverDnpotrsBatched()`](https://docs.nvidia.com/cuda/cusolver/index.html#cuSolverDN-lt-t-gt-batchpotrs), which is called with a batch of pairs `A` and `b`. For [`cusolverDnpotrs()`](https://docs.nvidia.com/cuda/cusolver/index.html#cuSolverDN-lt-t-gt-potrs), `nrhs` can be arbitrary (corresponding to the same matrix A and a batch of vectors b). Since `cupyx.lapack.posv()` supports both use cases, we should get it modified.

Open cupyx/lapack.py around line 274 and read the posv() docstring alongside the linked cuSolver potrs and potrsBatched references. Update the documentation so it accurately covers both supported use cases, then verify that the wording distinguishes arbitrary nrhs from batched pairs of A and b.

Written by the indexing model from the issue text.