feat: possession presentation layer — glitchy text + degraded audio

First sub-project of the "make contact feel real" arc (spec:
docs/superpowers/specs/2026-07-23-possession-presentation-design.md).
Direct Contact replies now feel like a spirit fighting through static to
hold the channel rather than a plain chat bubble:

- backend/app/possession.py: compute_stability(rarity, magnitude, rng) —
  a 0.05-0.98 score per reply (rarer entity + stronger triggering anomaly =
  cleaner signal), rng injectable for a later quantum-RNG source.
- ws.py sends stability on reply_start; audio synthesis for that reply gets
  noise/bitcrush scaled by instability (1 - stability) via a new
  instability param on synthesize_spirit_voice — effects.py itself is
  untouched, only the params fed into it.
- frontend/src/lib/possession.ts: renderPossessedText — pure, deterministic
  (tick-seeded, no Math.random) text corruption with self-correcting
  glitch bursts, wired into Transcript.tsx's streaming reply display.

Stored transcript/reply text is unaffected — this is presentation only.
78/78 backend, 137/137 frontend tests passing.

backend/app/tts/piper.py

@@ -50,18 +50,48 @@ class PiperTTS:
 
 
 async def synthesize_spirit_voice(
-    text: str, voice: Voice, voice_profile: dict | None = None
+    text: str,
+    voice: Voice,
+    voice_profile: dict | None = None,
+    instability: float = 0.0,
 ) -> bytes:
-    """One call: Piper synth + the spirit's signature effects chain."""
+    """One call: Piper synth + the spirit's signature effects chain.
+
+    `instability` (0.0-1.0, i.e. `1 - stability`) scales up the noise and
+    bitcrush fed into the effects chain when the spirit is fighting to hold
+    the channel — a shaky possession sounds shakier. It never touches
+    `apply_effects` itself, only the values handed to it here.
+    """
     profile = voice_profile or {}
+    instability = max(0.0, min(1.0, instability))
     model_path = Path(settings.piper_voices_dir) / voice.model_file
+
+    base_noise = float(profile.get("noise", 0.03))
+    # Noise ramps linearly across the whole range so degradation is audible
+    # from the first sign of instability, but doubling the base level at
+    # instability=1.0 (rather than, say, 5x) keeps the words intelligible
+    # even at a full-static possession — this is texture, not white noise.
+    noise_level = base_noise + instability * base_noise
+
+    base_bitcrush = int(profile.get("bitcrush", 0))
+    # Bitcrush only kicks in once instability crosses ~0.5 -- a slightly
+    # unstable channel just sounds noisier, not crunchy/robotic. Above that
+    # threshold it ramps up to +6 bits of crush on top of whatever the
+    # voice's own profile already applies, capped at 15 (apply_effects only
+    # crushes when 0 < bits < 16).
+    if instability > 0.5:
+        bitcrush_bonus = round((instability - 0.5) / 0.5 * 6)
+    else:
+        bitcrush_bonus = 0
+    bitcrush_bits = min(15, base_bitcrush + bitcrush_bonus)
+
     wav = await PiperTTS(str(model_path)).synthesize(text)
     return await asyncio.to_thread(
         apply_effects,
         wav,
-        noise_level=float(profile.get("noise", 0.03)),
+        noise_level=noise_level,
         pitch_semitones=float(profile.get("pitch", 0.0)),
         rate=float(profile.get("rate", 1.0)),
-        bitcrush_bits=int(profile.get("bitcrush", 0)),
+        bitcrush_bits=bitcrush_bits,
         echo=float(profile.get("echo", 0.2)),
     )