fix: prevent AMCP channel crash on Unicode/emoji in CG UPDATE (#1106)

[toontoet]

boost::log throws conversion_error when logging wide strings containing characters outside the system locale's encoding (e.g. emojis on C locale). Since this exception is unhandled in the AMCP protocol handler, it kills the connection when running CasparCG in Docker.

Fix by sanitizing log output with replace_nonprintable() in the three places where user-supplied wide strings are logged:

• AMCPProtocolStrategy: received AMCP messages
• html_producer: CEF console messages and renderer process messages
• strategy_adapters: sent response messages

Also replace boost::wformat with string concatenation in html_cg_proxy::update() to avoid a secondary crash path, and use explicit skip method for UTF-8 conversion to gracefully handle any invalid sequences.

This is a minimal rework of the earlier #1673 PR, addressing the review feedback to reduce code changes.

Test plan

• Send CG UPDATE with emoji content (e.g. {"message": "Hello 😀 World"}) — connection stays alive
• Send complex multi-byte emojis (flags 🇳🇱, families 👨‍👩‍👧‍👦, skin tones 👋🏻) — no crash
• Send mixed ASCII + Unicode + emoji content — works correctly
• Verify emojis render correctly in the HTML template
• Verify AMCP connection remains stable after multiple emoji updates

Test template: debug.html
Test script: emoji_utf8_test.js (use: node emoji_utf8_test.js localhost 5250)