fix: allow multiple API keys for OpenAI/Anthropic compatible providers

[hodtien]

Summary

• Remove the one-connection-per-node guard for OpenAI Compatible and Anthropic Compatible providers so each endpoint can hold multiple API key connections.
• Add explicit node-type validation so a compatible provider id must map to a node of the matching type (rejects misconfigured nodes with 400 instead of silently storing wrong data).
• Pass defaultModel when bulk-adding API keys for compatible providers (single-add already required it; bulk previously created connections with defaultModel: null).
• Remove stale "only one connection" copy from the unused page.new.js.

Custom Embedding nodes intentionally keep the single-connection guard (out of scope).

Why

Adding a second API key to a custom Anthropic/OpenAI Compatible provider returned Only one connection is allowed for this Anthropic Compatible node. Each API key is modeled as a separate providerConnection under one providerNode, so the guard was the only thing preventing multi-key endpoints. Removing it (and keeping node existence + type checks) enables multiple keys while preserving routing, priority, active/test status, and per-key model defaults.

Test plan

• npx vitest run --config tests/vitest.config.js tests/unit/compatible-provider-connections.test.js — 4 passed (multi-key success, node-type mismatch rejection, existing single-connection cases)
• Full unit suite shows no new regressions (24 pre-existing failures present on base master before this branch; this branch has the same set, none in touched files)
• Manual: add two API keys to one compatible endpoint via UI (single + bulk), confirm both appear and route