refactor: apply conventions across session, permission, tool, and util

[hakula139]

Summary

Session, permission, tool, and utility code repeated state handling and used inconsistent layouts and test conventions. Shared helpers now own those operations, every session-listing caller uses the typed list_paged(limit, scope) API, and first-prompt titles respect their terminal-column cap for CJK text.

• Session: group actor batch state, centralize request acknowledgments, walk transcripts through one owner, share sanitization ID collection, derive the persistence byte limit and failure message from one MiB constant, and build listing rows once with named column widths. Consolidate redundant chain and shutdown tests while preserving their behavioral coverage.
• Permission: share lexical path normalization after checking relative working directories and search-walk decisions, simplify wildcard compilation, and remove references to unimplemented controls. Classifier work remains tracked in Use a cheap model classifier to decide ambiguous calls in auto mode #131.
• Tools: name create and overwrite modes, share search-root resolution and result markers, make stream trimming explicit, and align test names and fixtures with their contracts.
• Utilities and file tracking: share mutex recovery and generic plural selection without count conversions, extract snapshot verification, keep tracker types beside their implementations, and consolidate duplicate tests with exact assertions.

Not applied

• Multiline expected outputs already use indoc!.
• Resolver matching has only an all flag, so there is no all / archived pair to replace there.
• Read view tests call result_view directly. Their section banner now names that function.

Left for a follow-up

• The XDG base helper reaches configuration and client modules and remains deferred.
• The archive-notice split and resolver error capitalization remain deferred with their TUI and slash callers.

Test plan

• Permission regressions cover relative cwd traversal and search-walk agreement.
• Targeted file tracker, utility, tool, and session tests pass, including the CJK title regression and exact log truncation boundaries.
• cargo fmt --all --check, cargo build, and cargo clippy --all-targets -- -D warnings.
• cargo test: 2,635 unit tests (1 ignored) and 5 CLI tests pass. cargo llvm-cov passes with no uncovered changed executable lines locally.
• pnpm format, pnpm lint, and pnpm spellcheck.
• Pre-commit hooks pass. No snapshot or release-fixture changes.
• A real CLI process against a mock rejected request preserves the git branch and exact user message in the session.
• CI passes on Linux, macOS, and Windows. Codecov confirms that all modified coverable lines are covered.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!