Add 'limitations' field to AgentCard

## Summary

Proposal to add an optional `limitations` field to the `AgentCard` message in the A2A specification.

## Motivation

Currently, the `AgentCard` describes what an agent **can** do (via `description`, `skills`, `capabilities`), but there is no structured way to express what an agent **cannot** do or where it may underperform.

Exposing limitations is important for:
- **Agent-to-agent orchestration**: A calling agent can make better routing decisions if it knows the limitations of downstream agents (e.g., language support, file size limits, domain restrictions).
- **User trust & transparency**: Users can set appropriate expectations before interacting with an agent.
- **Safety & compliance**: Explicitly declaring limitations helps prevent misuse in unsupported scenarios.

## Proposed Change

Add a `repeated string limitations` field to the `AgentCard` message in `specification/a2a.proto`:

```protobuf
// Optional. A human-readable description of the agent's limitations,
// helping users and other agents understand what the agent cannot do
// or where it may underperform.
// Example: ["Does not support languages other than English",
// "Cannot process images larger than 10MB"]
repeated string limitations = 15;
```

This field is optional (empty list by default) and fully backward-compatible.

## Examples

```json
{
"name": "Recipe Agent",
"description": "Agent that helps users with recipes and cooking.",
"limitations": [
"Does not support languages other than English",
"Cannot process images larger than 10MB",
"Not suitable for professional nutritional advice"
]
}
```