Panic in syntect regex: OnigError(-5) memory allocation failure on large MCP tool responses

Bug Description

Claude Code crashes with a Rust panic in the syntect syntax highlighting library when rendering large or complex tool responses (e.g., MCP tool output containing decompiled Java source code with escaped Unicode sequences).

Error

thread '<unnamed>' (16132) panicked at /Users/hellja/.cargo/registry/src/artifactory.infra.ant.dev-7db23613d841872b/syntect-5.3.0/src/parsing/regex.rs:70:53:
regex string should be pre-tested: Error(OnigError(-5), fail to memory allocation)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Reproduction Steps

1. Connect an MCP server that returns Java source code snippets (e.g., decompiled JDK classes via JDTLS java/classFileContents)
2. Call a tool that returns ~50 lines of Java source including Javadoc with HTML entities (\u003cp\u003e, \u003ci\u003e, etc.)
3. Claude Code attempts to syntax-highlight the response and crashes

Environment

• OS: Windows 11 Enterprise 10.0.26200
• Shell: PowerShell / Git Bash
• syntect version: 5.3.0 (from error message)

Expected Behavior

Tool responses should render without crashing, even if syntax highlighting fails. A graceful fallback to plain text would be preferable.

Possible Cause

The Oniguruma regex engine (OnigError(-5)) fails to allocate memory when parsing complex content. This could be triggered by:

• Large inline source code in JSON tool responses
• Escaped Unicode sequences in Javadoc comments
• Deeply nested regex patterns in the Java grammar

A catch_unwind around the syntect highlighting call or a memory limit guard could prevent the panic from crashing the entire process.