feat: upload audio, get a private Swedish transcript, discuss it (#33, ADR-033) (#39)

Upload a recording (audio, or video with the camera off), get a private Swedish transcript, and chat about it.

- ADR-033: RLS-scoped private transcript store; ffprobe/ffmpeg intake; gateway transcription (iguana kb-whisper, berget fallback); in-process single-replica worker; chat over the transcript (local models only).
- ADR-034: optional two-engine mode (pianissimo + kb-whisper) reconciled by a LOCAL model; off unless TAPIR_RECONCILE_BASE_MODEL is set.
- ADR-035: real progress (phases, running/quiet/lost), phase logging, every judged spot stored, merged segments.
- Hardening from four adversarial reviews: errors logged and stored as kinds never text; per-engine deadlines; converted-length check; bounded queue; client hang-ups are not errors; strict local-alias check.
- Dockerfile: static ffmpeg/ffprobe 7.0.2 pinned by sha256 (#34). Built and inspected by hand; CI does not build the image on a PR, so its first CI build is this merge.

Uploads stay OFF until TAPIR_UPLOADS_ENABLED is set in the deployment.

Closes #33, #34, #42, #43, #44.

Co-authored-by: Mathias <mthbqv@gmail.com>

internal/web/uploads_disconnect_handler_test.go

@@ -0,0 +1,124 @@
+package web_test
+
+import (
+	"context"
+	"fmt"
+	"io"
+	"log/slog"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// A client that hangs up in the middle of an upload is routine, not a server fault. The
+// in-process recorder cannot see it (its context never cancels), so this talks to a real
+// listener (third adversarial review: over a socket every such case logged an ERROR and
+// half-close returned a 500, because the error page was rendered on a cancelled context).
+
+// body positions at which the client can hang up: inside the title field, inside the file
+// part's own headers, and mid-file. Every one must be handled alike (fourth adversarial
+// review: the title branch had no guard and still logged an ERROR).
+const (
+	inTitle   = "in the title"
+	inHeaders = "in the file part's headers"
+	midFile   = "mid-file"
+)
+
+func partialUpload(t *testing.T, addr, where string, hangUp func(*net.TCPConn)) {
+	t.Helper()
+	c, err := net.Dial("tcp", addr)
+	require.NoError(t, err)
+	tc := c.(*net.TCPConn)
+	const boundary = "xBOUNDARYx"
+	var body string
+	switch where {
+	case inTitle:
+		body = "--" + boundary + "\r\nContent-Disposition: form-data; name=\"title\"\r\n\r\nhalf a tit"
+	case inHeaders:
+		body = "--" + boundary + "\r\nContent-Disposition: form-data; name=\"file\"; filena"
+	default:
+		body = "--" + boundary + "\r\nContent-Disposition: form-data; name=\"file\"; filename=\"a.m4a\"\r\nContent-Type: audio/mp4\r\n\r\n" + strings.Repeat("a", 4000)
+	}
+	_, err = fmt.Fprintf(tc, "POST /uploads HTTP/1.1\r\nHost: x\r\nContent-Type: multipart/form-data; boundary=%s\r\nContent-Length: 200000\r\n\r\n%s", boundary, body)
+	require.NoError(t, err)
+	time.Sleep(50 * time.Millisecond) // let the server start reading
+	hangUp(tc)
+}
+
+func TestUploadPost_AClientThatHangsUpAnywhereInTheBodyIsLoggedAtInfoAndNeverAsAnError(t *testing.T) {
+	hangUps := map[string]func(*net.TCPConn){
+		"a half close":  func(c *net.TCPConn) { _ = c.CloseWrite(); time.Sleep(100 * time.Millisecond); _ = c.Close() },
+		"a full close":  func(c *net.TCPConn) { _ = c.Close() },
+		"a reset (RST)": func(c *net.TCPConn) { _ = c.SetLinger(0); _ = c.Close() },
+	}
+	for _, where := range []string{inTitle, inHeaders, midFile} {
+		for name, hangUp := range hangUps {
+			t.Run(where+", "+name, func(t *testing.T) {
+				app, sub := newUploadApp(t, audioOK, 200<<20)
+				buf := &lockedBuf{}
+				app.Log = slog.New(slog.NewTextHandler(buf, nil))
+				srv := httptest.NewServer(app.Router())
+				defer srv.Close()
+
+				partialUpload(t, strings.TrimPrefix(srv.URL, "http://"), where, hangUp)
+
+				require.Eventually(t, func() bool {
+					out := buf.String()
+					return strings.Contains(out, "upload abandoned") || strings.Contains(out, "upload rejected") || strings.Contains(out, "level=ERROR")
+				}, 3*time.Second, 20*time.Millisecond, "the server never finished handling the aborted upload")
+				time.Sleep(100 * time.Millisecond)
+				out := buf.String()
+				assert.NotContains(t, out, "level=ERROR", "a client hanging up is not a server fault:\n%s", out)
+				assert.Empty(t, sub.jobs)
+			})
+		}
+	}
+}
+
+// goneMidAccept is an Uploader during whose Accept the client disappears: the request's
+// context is cancelled, then Accept fails the way a read from a dead connection does.
+type goneMidAccept struct{ cancel context.CancelFunc }
+
+func (g goneMidAccept) Accept(context.Context, string, string, io.Reader) (string, error) {
+	g.cancel()
+	return "", fmt.Errorf("receive: %w", io.ErrClosedPipe)
+}
+
+func TestUploadPost_AnAbandonedRequestGetsA499NotAnImplicitEmpty200(t *testing.T) {
+	// If the context is cancelled but the connection can still be written to (a client that
+	// half-closed after sending), a missing status becomes 200 with an empty body: a false
+	// success (fourth adversarial review).
+	app, _ := newUploadApp(t, audioOK, 200<<20)
+	buf := &lockedBuf{}
+	app.Log = slog.New(slog.NewTextHandler(buf, nil))
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	app.Uploads = goneMidAccept{cancel: cancel}
+
+	rec := do(t, app, uploadReq(t, "t", "a.m4a", []byte("x")).WithContext(ctx))
+
+	assert.Equal(t, 499, rec.Code, "never an implicit 200 for an upload that was not accepted")
+	assert.Contains(t, buf.String(), "upload abandoned")
+	assert.NotContains(t, buf.String(), "level=ERROR")
+}
+
+func TestUploadPost_ADatabaseDropThatLooksLikeAnEOFIsStillAServerError(t *testing.T) {
+	// pgx turns a connection that drops mid-message into io.ErrUnexpectedEOF. Mapping that
+	// sentinel to "your upload was cut off" (a 400) would hide a real fault: only an error
+	// from reading the CLIENT's body is theirs (third adversarial review).
+	app, _ := newUploadApp(t, audioOK, 200<<20)
+	buf := &lockedBuf{}
+	app.Log = slog.New(slog.NewTextHandler(buf, nil))
+	app.Uploads = failingUploader{err: fmt.Errorf("failed to receive message: %w", io.ErrUnexpectedEOF)}
+
+	rec := do(t, app, uploadReq(t, "t", "a.m4a", []byte("x")))
+
+	assert.Equal(t, http.StatusInternalServerError, rec.Code)
+	assert.Contains(t, buf.String(), "level=ERROR")
+}