Reconsider clipping padding for geometry-dependent OOB cases

Following up on #6881, I wanted to clarify one limitation of handling this through oob.

The existing oob functions are very useful and solve many out-of-bounds problems, but none gives the desired result for a survival curve truncated at an arbitrary follow-up time. Censoring removes the first observation beyond the limit, so geom_step() cannot extend the final step to the boundary; squishing moves subsequent observations onto the boundary and can introduce unwanted steps there; and keeping the observations preserves the correct geometry but returns to the original clipping problem.

More importantly, this cannot be solved cleanly by simply adding a survival-specific x-scale oob function. To truncate a step curve correctly, the endpoint at the x-limit must retain the preceding y-value, and with multiple curves this must also be done within each group. That requires information about how the geom constructs its geometry, not just the out-of-bounds x-values available to the scale.

This illustrates a broader issue: some specialised geometries may require geom-specific handling of out-of-bounds observations. Relying on oob for these cases could therefore require separate solutions at different levels of the plotting pipeline rather than a general scale-level mechanism.

That is where a small optional clipping allowance could still be valuable as a last-resort mechanism. It would allow the full data to construct the correct geometry and then clip that geometry just outside the panel, without requiring specialised preprocessing or geom-specific handling for each case.

Would this be a reason to reconsider reopening #6881? I also completely understand if this kind of fallback is considered too specialised for ggplot2 itself and better suited to an extension package.

Start with issue #6881 and review how the existing oob handling, geom_step(), and x-scale behavior interact for truncated survival curves. Determine whether an optional clipping allowance is appropriate for geometry-dependent cases and what design would preserve grouped step endpoints; done means the project reaches a clear decision on reopening the proposal.