hooks/pyramid_attention_broadcast: remove redundant iteration==0 guard and fix stale cache VRAM leak

[GitGlimpse895]

What does this PR do?

Fixes two bugs in PyramidAttentionBroadcastHook.new_forward:

1. Redundant iteration == 0 condition
After every reset_state(), both iteration and cache are reset to 0 and None respectively. So at iteration 0, self.state.cache is None is always True, making self.state.iteration == 0 permanently dead code that creates a misleading impression of two independent invariants.

2. Stale cache leaking GPU VRAM
When outside the active timestep range, the hook was unconditionally writing self.state.cache = output, holding a full hidden-state activation tensor on GPU until the next generation's reset_state() call. For video transformers with dozens of PAB-hooked layers, this accumulates hundreds of MBs of unreleased VRAM. The fix sets self.state.cache = None immediately when outside the range.

This is a re-submission of #13467 (accidentally self-closed). Apologies for the noise, @sayakpaul!

Checklist

• Did you read the contributor guideline?
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

Who can review?

@sayakpaul @DN6