fix potensial bug: when n_importance not divided by up_sample_steps w…

[murumura]

Potential bug:
in renderer.py When self.n_importance is not divisible by self.up_sample_steps will cause line 365 in renderer.py reshape error due to shape differ.

For example when setting self.up_sample_steps=6 and self.n_importance=64 and self.n_samples=64 it will result tensor sdf in last iteration of up sample loop to have shape of (batch_size, 124) (since 64//6=10, we missing the non-zero remainder at each loop-iteration) instead of (batch_size, 128).

While in line 341 n_samples = self.n_samples + self.n_importance result 128 instead of 124, so this will make
s_val = ret_fine['s_val'].reshape(batch_size, n_samples).mean(dim=-1, keepdim=True) to cause a reshape error.

I simply add accumulative non-zero reminder to n_importance argument passed to self.up_sample at last iteration to resolve this bug.