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/uploads/intake_busy_test.go

@@ -0,0 +1,96 @@
+package uploads_test
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"io"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+
+	"git.d-ma.be/mathias/tapir/internal/domain"
+	"git.d-ma.be/mathias/tapir/internal/uploads"
+)
+
+// A full queue is the user's to retry, not a server fault: a typed rejection, the row
+// failed honestly, the audio removed (adversarial review F6).
+
+func TestAccept_AFullQueueIsABusyRejectionNotAServerErrorAndLeavesNothingBehind(t *testing.T) {
+	r := newRig(t)
+	r.sub.err = uploads.ErrQueueFull
+
+	_, err := r.in.Accept(context.Background(), "user-1", "t", strings.NewReader("data"))
+
+	var rej *uploads.RejectError
+	require.True(t, errors.As(err, &rej), "a full queue must be a rejection the web layer can answer with 429: %v", err)
+	assert.Equal(t, uploads.ReasonBusy, rej.Reason)
+	assert.Contains(t, rej.Message, "waiting")
+	assert.Equal(t, domain.UploadErrBusy, r.st.failed["up-1"], "the row must not sit queued: that is the stuck spinner")
+	assert.Empty(t, r.tmpFiles(t), "the audio is deleted")
+}
+
+// roomSubmitter is a Submitter that can say, before anything is spooled, that there is no room.
+type roomSubmitter struct {
+	fakeSubmitter
+	room bool
+}
+
+func (r *roomSubmitter) HasRoom(string) bool { return r.room }
+
+func TestAccept_ANoRoomQueueIsRefusedBeforeAnythingIsSpooledProbedOrStored(t *testing.T) {
+	// Checking only at Submit meant 200 MB was spooled and ffprobe run for every refused
+	// attempt, and a failed row was left behind each time (second adversarial review).
+	r := newRig(t)
+	full := &roomSubmitter{room: false}
+	in := uploads.NewIntake(r.st, r.pr, full, uploads.Limits{MaxBytes: 1024, MaxDuration: 90 * time.Minute}, r.tmp)
+
+	_, err := in.Accept(context.Background(), "user-1", "t", strings.NewReader("data"))
+
+	var rej *uploads.RejectError
+	require.True(t, errors.As(err, &rej))
+	assert.Equal(t, uploads.ReasonBusy, rej.Reason)
+	assert.Empty(t, r.st.created, "no row for an upload that was never accepted")
+	assert.Empty(t, r.pr.paths, "ffprobe was never run on it")
+	assert.Empty(t, r.tmpFiles(t), "nothing was spooled")
+	assert.Empty(t, full.jobs)
+}
+
+func TestAccept_ASubmitterThatCannotSayStillWorksAsBefore(t *testing.T) {
+	r := newRig(t)
+
+	_, err := r.in.Accept(context.Background(), "user-1", "t", strings.NewReader("data"))
+
+	require.NoError(t, err, "a Submitter without HasRoom is accepted: the check at Submit remains")
+}
+
+// brokenBody is a client body that fails mid-read, like a connection that dropped.
+type brokenBody struct{ n int }
+
+func (b *brokenBody) Read(p []byte) (int, error) {
+	if b.n > 0 {
+		b.n = 0
+		return copy(p, "partial audio"), nil
+	}
+	return 0, io.ErrUnexpectedEOF
+}
+
+func TestAccept_AnErrorReadingTheClientsBodyIsMarkedAsTheClientsAndAWrappedSubmitErrorIsNot(t *testing.T) {
+	r := newRig(t)
+
+	_, err := r.in.Accept(context.Background(), "user-1", "t", &brokenBody{n: 1})
+
+	require.Error(t, err)
+	assert.ErrorIs(t, err, uploads.ErrClientRead, "a dropped client body is the sender's to retry")
+	assert.Empty(t, r.tmpFiles(t))
+
+	// A fault that is ours must NOT carry the marker, even if it wraps the same sentinel.
+	r2 := newRig(t)
+	r2.sub.err = fmt.Errorf("db: %w", io.ErrUnexpectedEOF)
+	_, err = r2.in.Accept(context.Background(), "user-1", "t", strings.NewReader("data"))
+	require.Error(t, err)
+	assert.NotErrorIs(t, err, uploads.ErrClientRead)
+}