[epic] Unbounded, server-driven state growth

[ctoth]

Root cause: unbounded, server-driven growth

Multiple pieces of state can be driven without limit by the (untrusted) server or by normal long sessions, with no cap, eviction, or backpressure.

Instances already filed

• Auto-reconnect is unbounded and ignores server Goodbye/kick #59 auto-reconnect retries forever (no cap/backoff, ignores Goodbye)
• AutoLog pruneToMaxBytes can delete the session currently being written #57 autolog log growth / active-session deletion
• Music preload never hits cache (load/play use different keys); load has no eviction #69 MediaService sound cache has no eviction

Other observed (not separately filed)

• Telnet subnegotiation buffer grows without bound if IAC SE never arrives (src/telnet.ts)
• FileTransferManager.pendingOffers and the MCP multiline handler map have no per-sender cap

Suggested direction

Cap or evict every server-driven collection; add capped exponential backoff to reconnect; bound the subnegotiation buffer and error out past a limit.