refactor(agent): replace XML api_call with native function calling

Migrate the Telegram bot agent from an XML tag hack (<api_call>) to
OpenAI-native function calling via CallWithRequestFull.

Key changes:
- mcp/interface.go: add parseMCPResponseFull to clientHooks interface
- mcp/client.go: route callWithRequestFull through hooks for overridability
- mcp/claude_client.go: override parseMCPResponseFull for Claude response
  format (tool_use blocks instead of choices[].message.tool_calls)
- telegram/agent/agent.go: rewrite Run() to use CallWithRequestFull;
  define api_request tool with JSON Schema; implement tool-call loop
  with role="tool" result messages; remove XML parsing entirely
- telegram/agent/apicall.go: remove parseAPICall (dead code)
- telegram/agent/prompt.go: simplify — remove XML format instructions,
  replace with concise api_request tool usage instructions
- telegram/agent/agent_test.go: rebuild all tests using LLMResponse
  objects; add TestNarrationStructurallyImpossible, TestOnChunkCalledWithFinalReply,
  TestToolCallIDPropagated; remove XML-specific tests

Architecture advantage: with native function calling, the LLM returns
EITHER ToolCalls OR Content — never both. Narration is now structurally
impossible at the protocol level, not just enforced by prompt rules.

All 11 agent tests pass. mcp package tests pass.

mcp/request.go

@@ -1,9 +1,34 @@
 package mcp
 
-// Message represents a conversation message
+// Message represents a conversation message.
+// Supports plain messages (Role+Content), assistant tool-call messages (ToolCalls),
+// and tool result messages (Role="tool", ToolCallID, Content).
 type Message struct {
-	Role    string `json:"role"`    // "system", "user", "assistant"
-	Content string `json:"content"` // Message content
+	Role       string     `json:"role"`                  // "system", "user", "assistant", "tool"
+	Content    string     `json:"content,omitempty"`     // Text content (omitted when ToolCalls present)
+	ToolCalls  []ToolCall `json:"tool_calls,omitempty"`  // Set by assistant when calling tools
+	ToolCallID string     `json:"tool_call_id,omitempty"` // Set on role="tool" result messages
+}
+
+// ToolCall is a single function call requested by the LLM.
+type ToolCall struct {
+	ID       string           `json:"id"`       // Unique call ID (e.g. "call_abc123")
+	Type     string           `json:"type"`     // Always "function"
+	Function ToolCallFunction `json:"function"` // Function name and JSON-serialised arguments
+}
+
+// ToolCallFunction holds the function name and raw JSON arguments string.
+type ToolCallFunction struct {
+	Name      string `json:"name"`      // Function name
+	Arguments string `json:"arguments"` // JSON-encoded argument object
+}
+
+// LLMResponse is returned by CallWithRequestFull and carries both the assistant
+// text reply (Content) and any structured tool calls (ToolCalls).
+// Exactly one of the two fields will be non-empty for a well-formed response.
+type LLMResponse struct {
+	Content   string     // Plain-text reply (final answer)
+	ToolCalls []ToolCall // Structured tool invocations
 }
 
 // Tool represents a tool/function that AI can call