move context functions from agent/context.rs to thought/context.rs

trim_conversation moved to thought/context.rs where model_context_window,
msg_token_count, is_context_overflow, is_stream_error already lived.
Delete the duplicate agent/context.rs (94 lines).

Co-Authored-By: Proof of Concept <poc@bcachefs.org>

src/thought/context.rs

@@ -423,6 +423,45 @@ pub fn is_stream_error(err: &anyhow::Error) -> bool {
     err.to_string().contains("model stream error")
 }
 
+/// Trim conversation to fit within the context budget.
+/// Returns the trimmed conversation messages (oldest dropped first).
+pub fn trim_conversation(
+    context: &ContextState,
+    conversation: &[Message],
+    model: &str,
+    tokenizer: &CoreBPE,
+) -> Vec<Message> {
+    let count = |s: &str| tokenizer.encode_with_special_tokens(s).len();
+    let max_tokens = context_budget_tokens(model);
+
+    let identity_cost = count(&context.system_prompt)
+        + context.personality.iter().map(|(_, c)| count(c)).sum::<usize>();
+    let journal_cost: usize = context.journal.iter().map(|e| count(&e.content)).sum();
+    let reserve = max_tokens / 4;
+    let available = max_tokens
+        .saturating_sub(identity_cost)
+        .saturating_sub(journal_cost)
+        .saturating_sub(reserve);
+
+    let msg_costs: Vec<usize> = conversation.iter()
+        .map(|m| msg_token_count(tokenizer, m)).collect();
+    let total: usize = msg_costs.iter().sum();
+
+    let mut skip = 0;
+    let mut trimmed = total;
+    while trimmed > available && skip < conversation.len() {
+        trimmed -= msg_costs[skip];
+        skip += 1;
+    }
+
+    // Walk forward to user message boundary
+    while skip < conversation.len() && conversation[skip].role != Role::User {
+        skip += 1;
+    }
+
+    conversation[skip..].to_vec()
+}
+
 fn parse_msg_timestamp(msg: &Message) -> Option<DateTime<Utc>> {
     msg.timestamp
         .as_ref()