fix: batch song renamer and prepare for centralised song editing

[safepay]

Batch song renamer fixes and metadata extraction groundwork

Preparatory work for the metadata extraction plan outlined in #779. Cleans up and correctly structures batch_song_renamer.py so that its logic can be extracted into pikaraoke/lib/metadata_parser.py in a follow-up PR.

What changed

Format detection: artist-title and title-artist both handled correctly

The renamer now detects whether a filename uses Artist - Title or Title - Artist ordering and preserves that format in the suggested rename. Previous tests had the assertions backwards — this branch corrects both the logic and the tests.

Functions decomposed for testability

score_result was split into _score_query_match and _score_penalties. Helpers for accent removal, format detection, artist deduplication, and name comparison were extracted as named functions. This makes the scoring and format-detection logic independently testable and directly extractable into metadata_parser.py without refactoring.

Last.fm resilience

• Proactive rate limiting at 5 req/sec via _last_api_request_time
• Reactive retry with exponential backoff on HTTP 429 / API error 29
• _RATE_LIMITED sentinel distinguishes transient failures (don't cache) from genuine no-results (cache). Previously, timeouts and network errors were cached permanently as None.
• Plain dict cache replaces lru_cache to enforce this distinction

Case-only rename support

Two-step rename via a temp file handles case-insensitive filesystems (Windows, macOS) — e.g. renaming abba to ABBA no longer silently fails.

Route cleanup

• browse simplified to GET-only
• get_songs_to_rename query param song-index renamed to song_index (consistent with Python/schema conventions)

Test coverage added

• 9 tests for _detect_artist_first covering format detection edge cases
• 6 tests for rate limiting: retry behaviour, backoff timing, cache-vs-no-cache semantics
• Existing assertions corrected

Prepares for #779

The decomposed functions and plain-dict cache in this branch are the direct prerequisite for extracting them into pikaraoke/lib/metadata_parser.py. The follow-up will add regex_tidy() (lightweight cleanup for YouTube-sourced files), has_youtube_id() (provenance detection), and server-side endpoints to replace the client-side Last.fm calls in edit.html.