GUI & Storage modules + protocol fixes + singleton lock to prevent multi-session transport corruption

[thevan4]

What

⚠️ Second MCP server instance will exit immediately at startup — new Claude session will get a dead MCP server; singleton lock enforced (see Other for details) - this is a bit of a holy war change, open to debate)

New MCP tools

• modules/gui/ — screen capture (ASCII/PNG) and button input injection
• modules/storage/ — upload local files to SD card, list, mkdir, delete

Protocol fixes

• storage_write — rewrote with 512-byte chunking + has_next; sending the whole file as a single protobuf message doesn't work on Flipper firmware
• GUI screen capture — added drain loop after StopScreenStream; without it the next RPC call reads a stale response from the buffer and desyncs
• All operation timeouts 3s → 15s — 3 seconds wasn't enough for real workloads and slower connections

Other

• Singleton lock in CLI (/tmp/flipper-mcp.lock) — lock is OS-global, works across all projects and Claude sessions on the same machine; multiple MCP sessions pointing at the same Flipper will interleave RPC frames on the shared transport and corrupt the session; the lock ensures only one server process runs at a time
• Removed dead FlipperApp.stop() stub (always returned False), replaced with FlipperGUI
• MCP scope note — configuring MCP servers at user/global scope is generally NOT recommended; prefer project-scope to avoid polluting all sessions. Exception: flipperzero must be in user-scope — if configured per-project, the singleton lock still works OS-globally, but a second Claude window opening a different project will spin up its own instance and bypass the lock's intent. User-scope ensures exactly one server regardless of which project is active.

[thevan4]

@claude please review this PR

[thevan4]

4 MCP user-scope setup: claude mcp add --scope user --transport stdio flipperzero -- /path/to/.venv/bin/flipper-mcp - mb should this also go into the README?..