Fix Lua stack corruption in window:application() when process terminated

[delaneyb]

Summary

Fixes intermittent crashes with assertion: "not enough elements in the stack" when calling hs.window:application() on windows from recently-terminated applications.

Root Cause

• The binding used the LuaSkin shared singleton to push the return value. That singleton holds a mutable lua_State pointer (.L).
• At the crash point, the singleton's .L differed from the binding's L, so the return value was pushed onto a different lua_State than the caller was returning to.
• Evidence from runtime logs:
  • Before push: WINAPP before: L=0x600002ae2158, skin.L=0x15b00a208, app=nil
  • pushNSObject ran on L=0x15b00a208
  • After: WINAPP after: L=0x600002ae2158 top=0, pushed=1 → Lua asserted "not enough elements in the stack".
• The mutable .L can be changed by any code path using the shared singleton (e.g. logging or other sharedWithState calls). The logs show a concrete L mismatch at the moment app==nil; this change removes sensitivity to that for this binding.

The Fix

• When app == nil (process terminated), push nil directly to the caller's L using lua_pushnil(L) instead of pushNSObject (which pushes to the singleton's .L).
• Clear the caller's stack first with lua_settop(L, 0) to avoid leaking the receiver (consistent with other window methods).

Precedent in Codebase

• extensions/axuielement/common.m uses the same pattern:
  • if init returns nil → lua_pushnil(L); else → [skin pushNSObject:obj].

Impact

• Minimal change; only affects the nil-return path.
• No behavior change for valid applications.
• Eliminates the observed state-mismatch crash during window iteration when the owning process has just terminated.

[github-actions]

This pull request does not contain a valid label. Please add one of the following labels: ['pr-fix', 'pr-change', 'pr-feature', 'pr-maintenance']