fix(orchestrator): Remove duplicate message alternation in provider layer

**Problem:**
Messages were being merged TWICE - once in AgentOrchestrator.ensureMessageAlternation()
and again in GitHubCopilotProvider.enforceMessageAlternation(). This double-alternation
was over-merging messages and potentially losing context.

**Root Cause:**
Both the orchestrator and provider had their own alternation logic:
1. AgentOrchestrator line 173: ensureMessageAlternation (smart, preserves tools)
2. GitHubCopilotProvider line 1258: enforceMessageAlternation (simple merge)

All messages pass through orchestrator FIRST, then provider, causing duplicate processing.

**Solution:**
- Removed enforceMessageAlternation from GitHubCopilotProvider
- Kept ensureMessageAlternation in AgentOrchestrator (better logic + logging)
- Added comment explaining why provider doesn't do alternation

**Testing:**
✅ GitHub Copilot: Works correctly (no double-merge)
✅ Claude: Works correctly (single alternation only)
✅ Tested with alternation disabled - confirmed alternation NOT the cause of task quality issues

**Note:**
This fix resolves the double-alternation bug. Separate task quality issues with Claude
(incomplete task execution) are unrelated to alternation and require separate investigation.

Author: fewtarius

Sources/APIFramework/AgentOrchestrator.swift

@@ -4507,8 +4507,6 @@ public class AgentOrchestrator: ObservableObject, IterationController {
         /// Claude requires strict user/assistant alternation with no empty messages
         /// This MUST happen before YARN because YARN compresses individual messages
         /// If we merge AFTER YARN, we concatenate compressed content and blow up token count!
-        /// TEMPORARY: Force alternation to run to diagnose bug
-        /// TODO: Remove this after fixing ensureMessageAlternation
         messages = ensureMessageAlternation(messages)
         logger.debug("callLLMStreaming: Applied message alternation fix - \(messages.count) messages after merging")
 

Sources/APIFramework/Providers.swift

@@ -1480,15 +1480,11 @@ public class GitHubCopilotProvider: AIProvider, ObservableObject {
             return !isToolResultPreview
         }
 
-        /// MESSAGE_ALTERNATION FIX: Enforce strict user/assistant alternation
-        /// GitHub Copilot requires alternating roles (user → assistant → user → assistant)
-        /// Without this, consecutive same-role messages cause token counting errors
-        let alternationFixedMessages = enforceMessageAlternation(filteredMessages)
-        if alternationFixedMessages.count != filteredMessages.count {
-            logger.debug("MESSAGE_ALTERNATION: Applied fix - \(filteredMessages.count) → \(alternationFixedMessages.count) messages")
-        }
+        /// NOTE: Message alternation is handled in AgentOrchestrator.ensureMessageAlternation()
+        /// before reaching this provider. Doing it here would cause double-merging and lose context.
+        /// The orchestrator version preserves tool messages and has better logging.
 
-        let messages = alternationFixedMessages.map { message in
+        let messages = filteredMessages.map { message in
             var messageDict: [String: Any] = [
                 "role": message.role
             ]