Support native async tool calls to remove sync wrappers

[ajbozarth]

Follow-up to #1032.

Context

Mellea's tool-call dispatch (_acall_tools)
is already async. The sync-wrapping friction is one layer down: ModelToolCall.call_func()
and MelleaTool.run() are both sync. When a tool wraps an async callable — such as
the MCP tool callables proposed in #1042, or user async functions wrapped via
from_callable per #1041 — the async function has to be re-entered from sync code
via _run_async_in_thread.

Feature

Give MelleaTool a native async call path (e.g., an arun coroutine method and
an async-capable counterpart to ModelToolCall.call_func) so _acall_tools can
await the tool directly when its underlying callable is async.

Outcomes

• Async callables wrapped via MelleaTool.from_callable can be awaited without
  a thread hop; the _run_async_in_thread wrap added in feat: allow async functions as tools #1041 becomes unnecessary.
• The MCP module's _make_sync_call shim around _execute_tool (proposed in feat: add MCP tool integration #1042)
  becomes unnecessary; the async _execute_tool can be awaited directly.
• Sync tool callables keep working unchanged — the async path wraps them trivially.

Expected touchpoints

• MelleaTool / AbstractMelleaTool: new async call method.
• ModelToolCall.call_func: async counterpart, or make the existing one async.
• _acall_tools: await the tool instead of calling call_func() synchronously.
• MelleaTool.from_callable: drop the _run_async_in_thread wrap for async funcs
  once arun exists.
• mellea/stdlib/tools/mcp.py: drop _make_sync_call, expose the async closure
  through the new async path.

[ajbozarth]

I'm willing to take a stab at implementing this whenever we decide its a priority