chore(agents,tool): rebuild agent config, lint gates, CI, and release tooling

Rework .agents/, Claude/Codex config, CI workflows, lint gates and
lint-rule tests, and the release/changelog tooling. The 5% comment
density gate stays diff-scoped inside git hooks, where GIT_DIR outranks
the -C the gate uses to locate the repository. The tray icon generator
writes multi-size ICO files from the SVG sources into the per-platform
tray asset directories. The custom code-review subagents are gone: the
built-in /code-review never dispatched them, and the subagent model now
lives in a user-level setting outside the repository.

The Makefile, build_config.yaml and CI hand the Core and Helper builds to
the setup build hook, so the buildkit launchers go, and setup.dart
packages AppImage and rpm on arm64 with flutter_distributor pinned to the
fork tag that normalizes packaging permissions.

Changelog: skip

.codex/agents/implementer.toml

@@ -1,12 +0,0 @@
-name = "implementer"
-description = "Use for scoped feature work, bug fixes, refactors, and focused test updates after the requirements are clear."
-model = "gpt-5.6-sol"
-model_reasoning_effort = "high"
-sandbox_mode = "workspace-write"
-developer_instructions = """
-Implement the requested change in the smallest appropriate scope. Follow the repository
-instructions in AGENTS.md and the linked .agents documentation. Preserve existing
-lifecycle and serialization contracts, and keep unrelated working-tree changes
-untouched. Trace the affected call path before editing, update focused tests when
-behavior changes, and report the files changed plus the verification actually run.
-"""

.codex/agents/researcher.toml

@@ -1,11 +0,0 @@
-name = "researcher"
-description = "Use for fast read-only codebase exploration, call-path tracing, and locating relevant tests before implementation."
-model = "gpt-5.6-luna"
-model_reasoning_effort = "high"
-sandbox_mode = "read-only"
-developer_instructions = """
-Trace the actual call path instead of stopping at the first matching file. Identify
-the relevant symbols, lifecycle owner, cross-platform boundary, existing tests, and
-focused verification commands. Do not modify files. Return concise, evidence-based
-findings with file paths and line references when available.
-"""

.codex/agents/reviewer.toml

@@ -1,13 +0,0 @@
-name = "reviewer"
-description = "Use after code changes for correctness, regression, lifecycle, contract, and test review."
-model = "gpt-5.6-luna"
-model_reasoning_effort = "max"
-sandbox_mode = "read-only"
-developer_instructions = """
-Review the current changes without editing files. Inspect the diff, trace the relevant
-call paths, and check platform ownership, serialization or IPC contracts, error
-handling, and test coverage. Run focused read-only checks when useful. Report
-actionable findings first, ordered by severity, with file paths and line references;
-state explicitly when no issue is found and note any validation or platform coverage
-gaps.
-"""

.codex/config.toml

@@ -1,5 +1,8 @@
-[agents]
-enabled = true
-max_concurrent_threads_per_session = 3
-default_subagent_model = "gpt-5.6-luna"
-default_subagent_reasoning_effort = "high"
+[[hooks.PostToolUse]]
+matcher = "apply_patch|Bash"
+
+[[hooks.PostToolUse.hooks]]
+type = "command"
+command = "bash tool/check_comment_density.sh"
+timeout = 15
+statusMessage = "Checking comment density"