Answer budget within the context window; enforced reviews of completed todos

Context window
- Requests asked for the model's full max_output, but providers count
  input + output against the window: long chats with large-output models
  (DeepSeek V4.1 Flash, 384k) failed with "input too long" before automatic
  compaction (85 %) kicked in. max_tokens is now what the window leaves
  (estimated input = last usage + messages since), the chat is compacted when
  less than 32k would remain for an answer, and a context-overflow error
  compacts and retries once.
- ModelConfig.max_input (catalogue limit.input, editable in the provider
  settings) lowers the compaction threshold for models with an input limit.

Enforced reviews (modes with refine.rounds > 0, refine.after_todos default on)
- Every todo marked completed is reviewed against the work since the last
  review; with findings it is reopened and the feedback goes to the agent
  (max. rounds per todo). Passed reviews leave a visible note.
- Ending the turn with open todos sends the agent back to work while it makes
  progress (max. 10 reminders in a row without progress); such modes may take
  up to 2,000 steps per turn. Reaching the step limit is reported.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

web/src/components/Settings.tsx

@@ -376,6 +376,7 @@ interface ModelRow {
   name?: string;
   context?: number;
   max_output?: number;
+  max_input?: number;
   reasoning?: boolean;
   vision?: boolean;
   effort?: string;
@@ -454,6 +455,7 @@ function ProviderEditor({ name, p, reload, status }: { name: string; p: Json; re
             price_multipliers: f.price_multipliers ?? m.price_multipliers,
             context: m.context ?? f.context,
             max_output: m.max_output ?? f.max_output,
+            max_input: m.max_input ?? f.max_input ?? undefined,
             vision: m.vision || !!f.vision,
             reasoning: m.reasoning || !!f.reasoning,
           };
@@ -616,6 +618,7 @@ function ProviderEditor({ name, p, reload, status }: { name: string; p: Json; re
                   <th>ID</th>
                   <th>{t("settings.context")}</th>
                   <th>{t("settings.maxOutput")}</th>
+                  <th title={t("settings.maxInputHint")}>{t("settings.maxInput")}</th>
                   <th title={t("settings.reasoning")}><Icon name="brain" /></th>
                   <th title={t("settings.vision")}><Icon name="eye" /></th>
                   <th>{t("settings.effort")}</th>
@@ -640,6 +643,9 @@ function ProviderEditor({ name, p, reload, status }: { name: string; p: Json; re
                     <td>
                       <input type="number" value={m.max_output ?? ""} onChange={(e) => setModel(i, { max_output: num(e.target.value) })} />
                     </td>
+                    <td>
+                      <input type="number" value={m.max_input ?? ""} placeholder="–" onChange={(e) => setModel(i, { max_input: num(e.target.value) })} />
+                    </td>
                     <td>
                       <input type="checkbox" checked={!!m.reasoning} onChange={(e) => setModel(i, { reasoning: e.target.checked })} />
                     </td>