[FEATURE] Show deferred MCP tools overhead in /context command

Summary

The /context command doesn't show the potential overhead of deferred MCP tools, creating confusion when users see "Large MCP tools context" warnings elsewhere.

Problem

When running /context, users see a breakdown like:

Context Usage
claude-opus-4-5-20251101 · 27k/200k tokens (13%)

Estimated usage by category
⛁ System prompt: 3.5k tokens (1.7%)
⛁ System tools: 16.9k tokens (8.4%)
⛁ Custom agents: 1.9k tokens (1.0%)
⛁ Memory files: 1.2k tokens (0.6%)
⛁ Skills: 2.3k tokens (1.2%)
⛁ Messages: 1.1k tokens (0.6%)
⛶ Free space: 140k (70.0%)

Meanwhile, at startup or via /doctor, they see:

Context Usage Warnings
 └ ⚠ Large MCP tools context (~39,769 tokens > 25,000)
   └ MCP servers:
     └ Sentry: 14 tools (~10,174 tokens)
     └ atlassian: 28 tools (~8,980 tokens)
     └ playwright: 44 tools (~8,377 tokens)
     └ claude-in-chrome: 17 tools (~5,604 tokens)
     └ ...

The disconnect is confusing: 13% context used, but ~40k tokens of MCP tools lurking.

Why This Matters

• Users see low context usage but get warnings about large MCP overhead
• The deferred tools become active when used, suddenly consuming that 40k
• No single view shows both current usage AND potential overhead

Suggested Solution

Add a line to /context for deferred MCP tools:

Estimated usage by category
⛁ System prompt: 3.5k tokens (1.7%)
⛁ System tools: 16.9k tokens (8.4%)
⛁ MCP tools (deferred): ~39.8k tokens (potential)  ← NEW
⛁ Custom agents: 1.9k tokens (1.0%)
...

Or show the warning inline if it exceeds a threshold.

Related Issues

• #17959 - used_percentage doesn't include system overhead
• #21553 - context_window in hooks excludes system overhead
• #18400 - /context header/table percentage mismatch
• #12241 - Large MCP tools context warning

This is distinct: those issues are about already-loaded overhead being excluded. This issue is about showing potential overhead of deferred tools so users can understand the warning in context.