Readme logging and debugger

Co-authored-by: Sam Zanca <metruzanca@users.noreply.github.com>

- Remove LogManager class, use simple functions
- Log server: accept any HTTP method, silent responses
- Remove JSON parsing attempt - just append raw body
- Rename prompt from 'mcp-debugger-instructions' to 'debug'
- Condense prompt text
- Remove unused imports (zod)

Log server reduced from 110 to 54 lines
Index reduced from 218 to 106 lines

- Switch from debug-logs.json to .debug.log plaintext file
- Use appendFile instead of read-parse-write cycle
- Each log entry is one line (strings or JSON stringified)
- Much faster - no parsing needed for appending
- Simpler format for agents to consume

- Document that server accepts JSON or plain strings
- Show multiple logging examples (objects, strings, template literals)
- Simplify log format documentation
- Remove outdated HTTP metadata from examples

- Try JSON.parse first, fall back to raw string on failure
- Skip empty bodies to avoid noise in logs
- Return 500 only on actual write failures
- Supports flexible debugging: objects, arrays, strings, quick notes

Examples that now work:
- fetch('localhost:6969', { body: 'hi' }) -> logs 'hi'
- fetch('localhost:6969', { body: JSON.stringify({x:1}) }) -> logs {x:1}
- fetch('localhost:6969', { body: '[1,2,3]' }) -> logs [1,2,3]

- Remove HTTP metadata (method, url, headers) from log entries
- Log entries now just store the data object directly
- Simplify fetch calls in instructions to show clean JSON logging
- More developer-friendly format for debugging workflows

Changes:
- Updated LogManager type to use LogData instead of LogEntry
- Simplified LogServer to append raw data objects
- Updated MCP prompt instructions to reflect simplified format
- Improved overall debugging experience