Fix off-by-one in moving_wall_bc ghost fill (stegoton_1d)

[bypire]

Fixes #717.

moving_wall_bc fills the lower ghost cells with for ibc in range(num_ghost-1),
which writes ghost indices num_ghost-1 down to 1, so the outermost lower ghost
cell (index 0) never gets set. range(num_ghost) lets ibc reach num_ghost-1,
which writes index 0 too.

Quick check of just that loop (ghost cells marked NaN so an unwritten one shows up):

num_ghost = 2
  before  range(num_ghost-1): qbc[1, :2] = [nan -1.]
  after   range(num_ghost)  : qbc[1, :2] = [-2. -1.]

Before the change the outermost ghost cell is left unset; after, all num_ghost
cells are filled (the reflection pairs index num_ghost-ibc-1 with interior
num_ghost+ibc).

[bypire]

Heads up on the failing checks: they're the stegoton_1d regression tests, which
compare the example's final strain against a stored expected_sols.npy at L1 < 1e-6.
The fix changes that output (L1 ~ 2e-3 for classic, ~2e-2 for sharpclaw), so the
outermost lower ghost cell does feed into the stencil and the stored reference was
generated with the old behavior.

So expected_sols.npy needs regenerating with the fix in place. Happy to do that
and push it here if you'd like, or if you'd rather double-check the boundary
behavior first and regenerate it on your side, that's fine too.

[ketch]

Thanks! Yes, please go ahead and push an updated reference solution.

[bypire]

Pushed the regenerated expected_sols.npy (all four configs; the stegoton tests pass against it locally). Thanks!

[coveralls]

coverage: 47.07%. remained the same — bypire:fix-stegoton-ghost-717 into clawpack:master