Clamp savedY to valid bounds on terminal resize

[buzzboxmedia]

Summary

• Buffer.resize() clamps savedX to newCols - 1 but does not clamp savedY, allowing it to exceed the new row count after resize
• Terminal.resizeBuffers() can produce a negative savedY when recomputing the offset after resize (normalBuffer.y + dy where dy was captured before resize)
• Both cases cause abort() in the Buffer.y setter on DEBUG builds when cmdRestoreCursor assigns buffer.y = buffer.savedY

Fix

• Clamp savedY to [0, newRows - 1] in Buffer.resize(), matching the existing savedX clamp
• Clamp savedY to >= 0 in Terminal.resizeBuffers() after recomputing the offset

Test plan

• Verify terminal resize no longer crashes when a saved cursor position exceeds new row count
• Run existing SwiftTerm test suite

🤖 Generated with Claude Code

[migueldeicaza]

Mhm, I think that this has shown that there might be a bit of a misuses with yDisp, I think this might need a deeper fix.

[migueldeicaza]

I think that this called into question a much bigger problem, which is that currently the code mixes savedY as either a viewport row (0..<rows) and sometimes an absolute buffer row (yBase-based) - and with these differences what you found was a step in the right direction, but not complete.

I compared this to xterm.js which was the original inspiration, and they do not have that discrepancy - either it got fixed there, or I introduced the bug.

This patch:

https://gist.github.com/migueldeicaza/fb7470de4b4c9501c7eb660bc56a3ede

Now treats savedY as absolute (yBase-baesd). Also, it now converts absolute savedY back to viewport row (savedY - yBase) and then clamps via restrictCursor().

While auditing I found another difference, softReset now also resets the saved cursor position

Would you mind taking this for a spin?

[migueldeicaza]

I created a branch absolute-savedY, I need to test that on iOS as well, but I am tied up for the next couple of days.