Fix the run_tui name collision that broke luminary map --tui (and CI mypy)

[rossry]

What

luminary/mapping/store.py -> records.py (af9c574) also renamed run_tui's parameter to saved — which was already the local holding termios.tcgetattr(fd):

def run_tui(core: SessionCore, saved: MappingStore, fps: float) -> None:
    ...
    saved = termios.tcgetattr(fd)     # the records are gone from here on
    ...
    saved.save_state(core.state, core.plan)   # AttributeError: 'list' object ...

So luminary map --tui raised before the operator's first keystroke, on the resume refresh. mypy caught it; CI has been red since.

• parameter is now records (it is a MappingStore, from records.py)
• the terminal flags are tty_attrs

Why it got through

Every existing test monkeypatches run_tui away, so nothing had ever executed its body — the loop was covered only by parse_keys unit tests either side of it.

This adds one that runs the real loop over a pty and asserts both halves of the collision: the resume refresh reaches the records, and cbreak is undone on exit. Two traps if it ever needs editing, both documented inline — tty.setcbreak flushes pending input (TCSAFLUSH), so a keystroke written before the loop starts is discarded; and in canonical mode an unterminated one is never readable at all. A small typist thread presses q until the loop takes it.

Checks

• CI mypy target: Success: no issues found in 56 source files (was 2 errors)
• black --check: clean
• full suite: 439 passed

Not covered: the fix is verified by test and by types, but I have not run luminary map --tui against the sphere.