…230)

* feat: unify cache paths via platform-aware dirs::cache_dir() (#229)

Route every `all-smi` cache writer — record output, energy WAL, and the users-tab CSV export — through a single platform-aware helper built on `dirs::cache_dir()`, retire the three hard-coded `~/.cache/all-smi/...` literals, and delete the hand-rolled `cache_dir_for_all_smi()` reimplementation in `src/view/event_handler.rs`.

The new `crate::common::paths::cache_dir()` helper mirrors the existing `config_dir()` so the cache layer is internally consistent with the config layer. `record.output_dir` and `energy.wal_path` now store `Option<String>` — `None` means "use the platform helper", `Some(s)` is an explicit operator override resolved through `expand_tilde`. Consumers (`record::RecorderOptions::from_args_with_settings`, `api::server::run_api_mode`) call the helper when no override is present.

Adds `crate::common::cache_migration::migrate_legacy_cache_paths`, called from `main()` before any cache file is touched. On Linux without `$XDG_CACHE_HOME` the old and new paths are identical and it is a no-op; on macOS, Windows, and Linux with `$XDG_CACHE_HOME` set, it best-effort-renames `~/.cache/all-smi/{records,energy-wal.bin}` into the new root only when the destination is empty (never overwrites) and refuses to follow a pre-planted symlink at the source (same threat model as the writer-side `O_NOFOLLOW` defences). The users-CSV exports are deliberately not migrated — they carry per-second timestamps and bulk-renaming hundreds of files would be more disruptive than helpful.

Tests added: `paths::tests::cache_dir_ends_with_app_name`, `record::tests::resolve_output_no_cli_no_config_uses_platform_cache_dir`, and a five-test `cache_migration::tests` module (`try_move_skips_when_source_missing`, `try_move_renames_file_when_dst_missing`, `try_move_renames_directory_when_dst_missing`, `try_move_no_op_when_dst_exists`, `try_move_refuses_symlink_source`). Existing `record::tests` resolver tests updated for the `Option<String>` shape; `config::tests::energy_config_defaults_match_issue_spec` and the `ALL_SMI_ENERGY_WAL_PATH` env test updated to assert `None` / `Some(...)`.

Docs: README updates the `[energy] wal_path`, `[record] output_dir`, `--output` and users-CSV export rows; manpage updates the FILES and ENERGY ACCOUNTING sections; the `record --output` `long_help` and the example `config.toml` text spell out the per-platform cache locations. The v0.21.0 changelog gets a closing sentence describing the unification and the migration. The `config print` / `config print --json` renderer now shows the resolved on-disk path when no override is set so operators can confirm at a glance where data will land.

Closes #229. Builds on #227 (Scope A of #223). The writer-side hardening (`O_NOFOLLOW`, mode `0600`, symlink refusal at rotation) lives in `src/record/writer.rs`, `src/metrics/energy_wal.rs::open_secure_append`, and `src/view/event_handler.rs::open_export_secure` and applies unchanged to the new resolved paths. `ALL_SMI_ENERGY_WAL_PATH` and `ALL_SMI_RECORD_OUTPUT_DIR` env-var support and the TOML key names are preserved.

* docs: align ENERGY_HELP wal-path text with per-platform defaults

The ALL_SMI_ENERGY_WAL_PATH help line still showed the Linux-only
~/.cache/all-smi/energy-wal.bin default, while the man page, README
footnote, and RecordArgs long_help already describe the per-platform
resolution. Use <platform cache dir>/all-smi/energy-wal.bin instead so
all four doc surfaces describe the same default for #229.

* refactor: tighten whitespace handling in cache-path resolvers (#229)

Apply the two LOW defense-in-depth consistency fixes flagged by the security checker:

**L1 — record resolver (src/record/mod.rs):** Add .map(str::trim) before .filter(!is_empty()) on the configured output_dir chain. A TOML value of "   " (spaces only) now falls through to the platform cache helper instead of being honored as a literal relative directory, matching how resolve_wal_path() already handles whitespace-only wal_path values via str::trim.

**L2 — TOML loader (src/common/config_apply.rs):** apply_file_energy and apply_file_record now trim and reject whitespace-only strings before writing to settings.energy.wal_path and settings.record.output_dir respectively, matching the existing behavior of apply_env_record which already calls trim().is_empty(). The collapsed let-chain form (if let Some(x) = ... && !x.trim().is_empty()) satisfies clippy's collapsible_if lint.

**Tests added:**
- src/record/mod.rs: resolve_output_whitespace_only_config_dir_falls_through — asserts a whitespace-only output_dir does not leak into the resolved output path.
- tests/config_file_integration_test.rs: whitespace_only_wal_path_and_output_dir_treated_as_unset — loads a TOML with wal_path = "   " and output_dir = "   " and asserts both fields resolve to None.

**Docs:** Changelog entry for the Users tab updated to reference the platform-aware cache path (via [^cache] footnote) instead of the Linux-only literal ~/.cache/all-smi/users-<timestamp>.csv.

All targeted tests pass (1172 unit, 15 integration), clippy -D warnings clean, cargo fmt --check clean.

Refs #221, #222, #224, #225, #226, #227.

Align config path active discovery with the implicit loader's first-existing-candidate semantics, correct record output help and resolver docs against the binary default, and make the swap-color regression test derive crossterm's current Red SGR sequence instead of hard-coding older encodings.

Refs #221, #222, #224, #225, #226, #227.

Align config path active discovery with the implicit loader's first-existing-candidate semantics, correct record output help and resolver docs against the binary default, and make the swap-color regression test derive crossterm's current Red SGR sequence instead of hard-coding older encodings.