Chat agent hangs / loops on tile fetch failures — add timeouts and UX improvements

[returnSGD]

Problem

When running PixelRAG locally, the chat agent can get stuck in a loop:

1. Search succeeds but tile images are not readable (too small, or network timeout)
2. Agent keeps retrying different tiles, then falls back to non-existent web search/fetch tools
3. Each request can take 3+ minutes before completing

Root causes

• No fetch timeouts on API calls — slow networks cause indefinite hangs
• Default API URL uses port 30001, which is blocked by many firewalls (HTTPS port 443 works)
• maxTurns: 20 allows excessive retry loops
• System prompt doesn't guide the agent to stop after failed tile reads
• Chat page lacks a Lightbox — clicking tiles navigates to raw tile URLs with no way back
• crypto.randomUUID() crashes on LAN IP access (non-secure context)

Fix (PR #49)

• 30s timeouts on both search and tile fetch calls
• Default URL changed to https://api.pixelrag.ai
• maxTurns reduced to 8, prompt updated to limit retries
• Lightbox with Back button added to chat page
• crypto.randomUUID() fallback for non-localhost origins

Related

• PR: fix: add fetch timeouts, improve UX (back button, chat lightbox), prevent agent loops #49

[aafaq-rashid-comprinno]

All fixes from this issue were merged in PR #49 (30s timeouts, maxTurns=8, crypto.randomUUID fallback, default URL to https://api.pixelrag.ai). The current agent-server.mjs on main has all of these in place. This can be closed.