fix(ego-lint): skip signal callback bodies in constructor init check

[ZviBaratz]

Summary

• Add filter_signal_callbacks() to check-init.py that excludes signal callback body lines (.connect(), .connectObject()) from the constructor init-time check, while still checking the .connect() line itself (the receiver may be a Shell global accessed at construction time)
• Add test fixture init-constructor-signal@test with both .connect() and .connectObject() callbacks containing Shell globals, plus a direct Shell global access that should still be flagged
• Mark tiling-shell annotation for constructor signal callback FP as fixed

Context

Issue #118 identified 8 FPs for init/shell-modification across tiling-shell and v-shell. Research showed that 6 of 8 were already resolved by prior PRs (#21 scoped constructor checks to extension.js, #88 added arrow fn exemption). The remaining 2 (tiling-shell globalState.js constructor signal callbacks) are addressed by this PR.

Signal callbacks are deferred — they don't execute during construction, only when the signal fires at runtime. The .connect() line itself is kept because the receiver object (e.g., Main.panel.connect(...)) IS accessed immediately during construction.

Test plan

• New fixture init-constructor-signal@test verifies:
  • Direct Shell global in constructor is flagged (line 10)
  • .connect() callback body is NOT flagged (lines 15-17)
  • .connectObject() callback body is NOT flagged (lines 21-24)
• All 653 existing test assertions pass (0 regressions)
• Existing init fixtures unchanged: init-time-safety, init-helper-constructor, init-modification, constructor-gobject

Closes #118

🤖 Generated with Claude Code

[ZviBaratz]

Review: Approve ✓

Clean, well-scoped fix. The semantic reasoning is correct — signal callbacks are deferred, not executed at construction time — and the implementation is consistent with existing brace-depth tracking patterns in the file.

What's good:

• filter_signal_callbacks() correctly keeps the .connect() line (receiver IS evaluated at construction time) while filtering callback body lines
• Test fixture covers both .connect() and .connectObject(), plus a direct Shell global access that should still be flagged
• No regressions — existing init fixtures unaffected

Minor observations (non-blocking):

1. Multi-line .connectObject() with callback on next line — if the call is split so the () => { starts on a different line than .connectObject(, the brace count on the .connect() line is 0, so callback_depth stays 0 and the callback body won't be filtered. Conservative behavior (flags when uncertain), which is the right default
2. Single-line inline callbacks — this.connect('notify', () => Main.panel.x) keeps the whole line, so Main.panel gets flagged even though it's deferred. Unlikely in practice inside constructors
3. .connect_after() not matched by the regex — very rare, fine to address if/when it causes a real FP