[upstream #16998] #16998: Make the in-memory TTS cache size configurable #326

Local review finding on LibreChat-AI#16998 (local:e4601b3:L-001, recorded in the reviewctl ledger; not a GitHub thread).

LibreChat-AI#16998 keeps read-aloud audio in an in-memory cache per tab, capped at a fixed 50 recordings with least-recently-used eviction (MAX_ENTRIES in client/src/utils/ttscache.ts). Per the repository guidance that new limits ship configurable, the cap could become a librechat.yaml setting, for example under speech, defaulting to 50. A byte budget may suit clients with tight memory better than an entry count.

The fixed cap was chosen deliberately for LibreChat-AI#16998 to keep that redesign small.