fix(mcpplugin): use create_mcp_http_client for streamable HTTP transport

[sharath-tursio]

Problem

When headers are provided to create_streamable_http_transport (e.g. an
Authorization: Bearer token for authenticated MCP servers), the transport
creates a bare httpx.AsyncClient(headers=...) which inherits httpx's default
timeout of 5 seconds for all operations.

This causes MCP tool calls to silently time out on any backend that takes longer
than 5 seconds to respond — which is common for LLM-backed or database-backed
MCP servers. The request hangs with no error, and the tool result is never
returned to the agent.

Notably, when no headers are passed, the code falls through to
streamable_http_client(url) which internally calls create_mcp_http_client()
— giving a 30s connect timeout and 300s read timeout. So unauthenticated
servers work fine, but authenticated ones silently fail.

Fix

Replace the conditional httpx.AsyncClient / streamable_http_client branches
with a single call using create_mcp_http_client(headers=...), which applies
the correct MCP-recommended timeouts consistently regardless of whether headers
are present.

Before

if resolved_headers:
async with httpx.AsyncClient(headers=resolved_headers) as http_client: # 5s timeout!
async with streamable_http_client(url, http_client=http_client) as (r, w, _):
yield r, w
else:
async with streamable_http_client(url) as (r, w, _): # 300s read timeout
yield r, w

After

async with create_mcp_http_client(headers=resolved_headers or None) as http_client:
async with streamable_http_client(url, http_client=http_client) as (r, w, _):
yield r, w

Impact

• No breaking changes — behaviour is identical for unauthenticated servers
• Authenticated MCP servers (Bearer token, API key headers, etc.) now get the same 300s read timeout as unauthenticated ones
• Simplifies the code by removing the if/else branch