Implement contextual auto-approval in MCP servers for OSE perspective

## Context
From the retro on PR #708, we identified opportunities to reduce "helicopter approval" patterns and elevate to an OSE (Outside and Slightly Elevated) perspective.

## Problem
Currently, agents request approval for many routine operations that could be safely automated, creating unnecessary friction and preventing focus on higher-level decisions.

## Proposed Solution (Scoped to GitHub MCP Server)
Implement contextual auto-approval logic in the GitHub MCP server as a first step. Since we control these servers, we can make them inherently safer while enabling auto-approval.

### Scope for Initial Implementation
Focus on GitHub operations only:
- Auto-approve operations on private repos owned by the authenticated user
- Use `get_me()` to determine authenticated user (portable approach)
- Maintain manual approval for exceptions (delete repo, make public, transfer ownership)

### Example Implementation Pattern
```python
# On startup
AUTHENTICATED_USER = get_me()['login']

# Before operations
if repo_owner == AUTHENTICATED_USER and repo_private and operation not in EXCEPTIONS:
return execute()
else:
return request_approval()
```

## Benefits
- Immediate 50% improvement vs waiting for perfect solution
- Reduces friction for routine operations
- Maintains safety through contextual logic
- Allows human focus on strategic decisions
- Implements safety at tool level rather than permission level
- Aligns with OSE perspective from systems-stewardship principle

## Future Expansion
After GitHub implementation proves successful, can expand pattern to:
- Git MCP server
- Filesystem MCP server
- Other MCP servers based on context

Start by locating the GitHub MCP server implementation and its get_me() entry point, then map the repository-owner, privacy, and operation checks described in the issue. Done means routine operations on authenticated-user private repositories can proceed contextually while delete, publicizing, and ownership-transfer operations still request approval, with the existing approval behavior preserved elsewhere.

Written by the indexing model from the issue text.