Stage 2b: prepare, commit and recover renderer frames

Upstream's renderer emits while it thinks, so the frame is recorded off-lock
and replayed under one transaction only after its geometry, output, owner and
terminal generations still match. A managed write, resize, owner change or
handoff in between retires the batch without emitting a byte of it.

Discarding the output is only half of discarding the frame. The renderer has
already advanced: _last_screen became a baseline for a frame the terminal never
received, and the mode flags latch beside their emission, so a full repaint
never re-emits them. Recovery drops the baseline and re-establishes a small
enumerated state contract physically, then tells the renderer what is true --
not a snapshot of upstream fields, and not upstream's reset(), which emits
operations of its own and rewrites available-height bookkeeping.

Cursor position replies are correlated by order, since the wire carries no
generation, and a row inside the reserved band is rejected before it can set a
height that upstream would have computed as zero or negative.

The contract tests record why each of those choices is necessary against
prompt_toolkit 3.0.53, so an upgrade that changes them fails loudly.

Author: tleonhardt

cmd2/output_recorder.py

@@ -89,7 +89,9 @@ def capture(cls, output: Output) -> "PreflightFacts":
             rows_below = None
         try:
             descriptor: int | None = output.fileno()
-        except (io.UnsupportedOperation, AttributeError, OSError):
+        except (io.UnsupportedOperation, NotImplementedError, AttributeError, OSError):
+            # Backends disagree on how to say "no descriptor": Vt100 lets StringIO raise
+            # UnsupportedOperation, DummyOutput raises NotImplementedError outright.
             descriptor = None
         return cls(
             size=output.get_size(),