Store chat images separately; JPEG for large PNGs; gzip/brotli responses

Images of messages (pasted images, screenshots, generated images) were base64
inside the message JSON, so opening a chat downloaded all of them at once
(66 MB for a chat with 120 screenshots). They now live in a `blobs` table,
addressed by SHA-256; messages keep a reference (blob, size). The UI loads
them from /api/blobs/<hash> lazily with an immutable cache header, the runner
fills the bytes in before model requests, JSON exports stay self-contained.

- Store::add_message moves inline images out; large opaque PNGs become JPEG
  (q85) when smaller. Older chats are converted once at start-up in the
  background, unused images are removed (also after deleting a chat) and the
  database is vacuumed when that frees 32 MB or more.
- Responses are compressed with gzip/brotli (not images, SSE streams or
  binary downloads).

Example (copy of production): chat load 66 MB -> 1.5 MB (370 KB brotli),
database 90 MB -> 27 MB.

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

server/src/api.rs

@@ -80,6 +80,7 @@ pub fn router(app: AppState) -> Router {
         .route("/sessions/{id}/worktree", get(worktree_status).post(worktree_action))
         .route("/sessions/{id}/tabs", get(session_tabs).put(put_file_tabs))
         .route("/sessions/{id}/export", get(export_session))
+        .route("/blobs/{hash}", get(get_blob))
         .route("/sessions/{id}", get(get_session).patch(update_session).delete(delete_session))
         .route("/sessions/{id}/prompt", post(prompt))
         .route("/sessions/{id}/abort", post(abort))
@@ -181,9 +182,21 @@ pub fn router(app: AppState) -> Router {
         .route("/cli/install.sh", get(cli_install_script))
         .route("/cli/download/{target}", get(cli_download))
         .fallback_service(static_files)
+        .layer(compression())
         .with_state(app)
 }
 
+/// gzip/brotli for JSON, HTML, JS and CSS. Images and SSE streams are left alone (default
+/// predicate), and so are binary downloads (already compressed or large).
+fn compression() -> tower_http::compression::CompressionLayer<impl tower_http::compression::Predicate + Clone> {
+    use tower_http::compression::predicate::{DefaultPredicate, NotForContentType, Predicate};
+    let predicate = DefaultPredicate::new()
+        .and(NotForContentType::const_new("application/octet-stream"))
+        .and(NotForContentType::const_new("application/zip"))
+        .and(NotForContentType::const_new("application/gzip"));
+    tower_http::compression::CompressionLayer::new().compress_when(predicate)
+}
+
 fn cli_dir() -> PathBuf {
     std::env::var("CODECRAFT_CLI_DIR").map(PathBuf::from).unwrap_or_else(|_| PathBuf::from("/app/cli"))
 }
@@ -744,7 +757,8 @@ struct ExportQ {
 /// Download a session as Markdown (default) or JSON.
 async fn export_session(Extension(user): Extension<User>, State(app): State<AppState>, Path(id): Path<String>, Query(q): Query<ExportQ>) -> Result<axum::response::Response, ApiError> {
     let s = readable(&app, &user, &id)?;
-    let messages = app.store.messages(&id)?;
+    // JSON exports are self-contained: images inline instead of store references.
+    let messages = if q.format == "json" { app.store.messages_for_model(&id)? } else { app.store.messages(&id)? };
     let base: String = s.title.chars().map(|c| if c.is_alphanumeric() || c == '-' { c } else { '_' }).collect::<String>().trim_matches('_').to_string();
     let base = if base.is_empty() { format!("session-{}", &id[..8.min(id.len())]) } else { base };
     let (body, mime, ext) = if q.format == "json" {
@@ -879,9 +893,33 @@ async fn delete_session(Extension(user): Extension<User>, State(app): State<AppS
     crate::shells::kill_session(&app, &id);
     app.terminals.kill_session(&id);
     app.store.delete_session(&id)?;
+    // Images only this session used.
+    let a = app.clone();
+    tokio::task::spawn_blocking(move || {
+        if let Err(e) = a.store.gc_blobs() {
+            tracing::warn!("image cleanup: {e:#}");
+        }
+    });
     Ok(Json(json!({"ok": true})))
 }
 
+/// An image of a chat message (see blobs.rs). Addressed by the SHA-256 of its bytes,
+/// so it never changes and the browser may keep it.
+async fn get_blob(Extension(_user): Extension<User>, State(app): State<AppState>, Path(hash): Path<String>) -> Result<axum::response::Response, ApiError> {
+    if !crate::blobs::is_hash(&hash) {
+        return Err(not_found("image"));
+    }
+    let (mime, bytes) = app.store.blob(&hash)?.ok_or_else(|| not_found("image"))?;
+    Ok((
+        [
+            (axum::http::header::CONTENT_TYPE, mime),
+            (axum::http::header::CACHE_CONTROL, "private, max-age=31536000, immutable".to_string()),
+        ],
+        bytes,
+    )
+        .into_response())
+}
+
 #[derive(Deserialize)]
 struct PromptBody {
     text: String,