feat: exact signature-DB match for decoded KEY .sub files

A Flipper KEY file names an already-decoded protocol but carries no pulses.
Previously _decode_from_key ignored the signature database entirely: it
synthesized a fake signature from the category router's guess, so an exact,
known protocol scored like a fuzzy guess (~0.45) and surfaced the router's
coarser category (e.g. Princeton -> "Remote Control") while discarding the
real catalog manufacturer (e.g. HCS301 lost "Microchip").

Now a decoded protocol name is looked up (case-insensitive) in the signature
DB. On a hit we surface the REAL catalog signature -- true manufacturer and
category (Princeton -> "Garage Door Opener") -- with high, frequency-aware
confidence (0.95 when the band is consistent, 0.90 otherwise): an exact
database match, not a timing guess. Unknown names keep the router fallback so
uploads are still identified by family rather than silently dropped.

Advances the "database matching" + "confidence scoring (exact/partial)" goals.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

src/matcher/pattern_decoder.py

@@ -146,10 +146,41 @@ class PatternDecoder:
         Identify a decoded (KEY) .sub file from its Protocol name.
 
         KEY files have no RAW pulses but carry a decoded `Protocol:` name,
-        frequency and key. We route the protocol name to a device category and
+        frequency and key. Two cases:
-        emit a match surfacing the real Flipper protocol + routed category, so
+
-        decoded uploads are no longer silently unidentifiable.
+        1. The named protocol is in our signature database. This is the
+           strongest possible identification — the capture device already
+           demodulated and *named* the protocol, and we recognise it — so we
+           surface the REAL catalog signature (true manufacturer + category)
+           with high, frequency-aware confidence. This is an exact "database
+           match", not a timing guess, and it keeps the surfaced category
+           consistent with the catalog (e.g. Princeton -> Garage Door Opener)
+           instead of the router's coarser guess.
+
+        2. The name is unknown to us. Fall back to category routing so the
+           upload is still identified by family rather than silently dropped.
         """
+        known = self._lookup_known_protocol(metadata.protocol, metadata.frequency)
+        if known is not None:
+            freq_ok = bool(metadata.frequency) and known.matches_frequency(metadata.frequency)
+            # Exact, already-decoded match against a known protocol. Confidence
+            # is high but never a literal 1.0 (the key payload itself is not
+            # cryptographically verified); a consistent frequency lifts it.
+            confidence = 0.95 if freq_ok else 0.90
+            details = {
+                "predicted_category": known.category,
+                "source": "decoded_key_exact",
+                "bit_length": metadata.bit_length,
+                "frequency_match": freq_ok,
+                "manufacturer": known.manufacturer,
+            }
+            return [DeviceMatch(
+                protocol=known,
+                confidence=confidence,
+                match_method="decoded_key_exact",
+                details=details,
+            )]
+
         pred = self.category_router.route_by_protocol(
             metadata.protocol, metadata.frequency
         )
@@ -172,6 +203,30 @@ class PatternDecoder:
             details=details,
         )]
 
+    def _lookup_known_protocol(
+        self, name: Optional[str], frequency: Optional[int]
+    ) -> Optional[ProtocolSignature]:
+        """Exact (case-insensitive) protocol-name lookup in the signature DB.
+
+        When several catalog entries share a name, prefer one whose frequency
+        band matches the capture; otherwise return the first. Returns None if
+        the name is unknown.
+        """
+        if not name:
+            return None
+        target = name.strip().lower()
+        candidates = [
+            p for p in self.protocol_db.get_all()
+            if p.name.strip().lower() == target
+        ]
+        if not candidates:
+            return None
+        if frequency:
+            for cand in candidates:
+                if cand.matches_frequency(frequency):
+                    return cand
+        return candidates[0]
+
     def _identify_pulse_widths(self, pulses: List[int]) -> Tuple[int, int, int, int]:
         """
         Identify SHORT/LONG pulse and gap durations using robust timing analyzer