Figure.clear() leaves stale subplot state (subplotgrid returns dead axes)

[cvanelteren]

Pre-existing bug — reproduces identically on main, not caused by #698. Filing it because #698 is what gives this state a named owner (SubplotManager), which makes it the natural place to fix.

Repro

import ultraplot as uplt
fig, axs = uplt.subplots(nrows=1, ncols=2)
fig.clear()

print(fig.axes)                          # []          <- matplotlib cleaned up
print(list(fig._subplots.subplot_dict))  # [1, 2]      <- stale, dead axes
print(len(fig.subplotgrid))              # 2           <- hands them back to the user
print(fig.gridspec)                      # GridSpec(nrows=1, ncols=2)  <- stale
print(fig._subplots.counter)             # 2           <- never reset

Figure.clear() goes through matplotlib and empties fig.axes, but nothing resets the subplot bookkeeping. So after a clear():

• fig.subplotgrid returns axes that are no longer in the figure,
• fig._get_subplot(n) resolves to a dead axes,
• fig.gridspec still points at the old gridspec,
• the label counter keeps climbing.

(On main the same thing happens via fig._subplot_dict / fig._gridspec / fig._subplot_counter — this is a faithful description of long-standing behaviour, not something the refactor introduced.)

Suggested fix

Give SubplotManager a reset() that clears subplot_dict, zeroes counter, and drops _gridspec, and have Figure.clear() call it. Add a regression test — there is currently no test covering figure state after clear(), which is why this survived.