refactor(tests): simplify YAML fixture schema and test runner

YAML: rimosse op/description/score_name/assertions block — ora detect/process
come chiave diretta, assertions piatte sullo stesso livello del caso.

Runner: eliminato _run_assertions engine, assertions inline in test_preprocess.
Riduzione da ~170 a ~75 righe totali tra YAML + test.

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

tests/fixtures/preprocessors/cases.yaml

@@ -1,121 +1,71 @@
-# Preprocessor test cases — Step 1 (Local Agent V2)
+# Preprocessor test cases
 #
-# Schema per caso:
-#   id: "1.N"
-#   description: str
-#   score_name: str                     # nome score inviato a Langfuse
+# detect: <expected_type>   → chiama detect_content_type(filename, content)
+# process: <content_type>   → chiama preprocess(content_type, content)
 #
-# Sorgente contenuto (una delle due):
-#   file: <nome file in data/>          # letto come testo UTF-8
-#   generate: binary_noise              # contenuto generato dal runner (per test binari)
+# Sorgente: file: <nome in data/>  oppure  generate: binary_noise
+# filename: override del nome file passato a detect (default: valore di file:)
 #
-# Per op=detect:
-#   op: detect
-#   input_filename: str                 # filename passato a detect_content_type
-#   expected_content_type: str
-#
-# Per op=preprocess:
-#   op: preprocess
-#   input_content_type: str             # content_type passato a preprocess()
-#   assertions:
-#     no_html_tags: bool
-#     min_length: int
-#     compression_ratio_lt: float       # len(clean) / len(raw) < soglia
-#     metadata_keys: [str, ...]         # chiavi che devono essere in metadata
-#     contains: str | [str, ...]        # substring(s) presenti in clean_text
-#     not_contains: str | [str, ...]    # substring(s) assenti da clean_text
-#     content_type: str                 # valore atteso di result.content_type
+# Assertions piatte (solo per process):
+#   no_html: true           clean_text senza tag HTML
+#   min_chars: N            len(clean_text) >= N
+#   ratio_lt: F             len(clean) / len(raw) < F
+#   has_meta: [k, ...]      chiavi presenti in metadata
+#   contains: str | [str]   substring(s) presenti in clean_text
+#   excludes: str | [str]   substring(s) assenti da clean_text
+#   content_type: str       result.content_type == questo valore
 
-cases:
+- id: "1.1"
+  file: email_action.html
+  filename: email_export.html
+  detect: email_html
 
-  # ── Detection tests ────────────────────────────────────────────────
+- id: "1.2"
+  file: generic_page.html
+  filename: index.html
+  detect: generic_html
 
-  - id: "1.1"
-    description: "Detect email HTML"
-    score_name: preprocess.detect_email
-    file: email_action.html
-    op: detect
-    input_filename: email_export.html
-    expected_content_type: email_html
+- id: "1.3"
+  file: notes.txt
+  detect: plain_text
 
-  - id: "1.2"
-    description: "Detect generic HTML"
-    score_name: preprocess.detect_generic
-    file: generic_page.html
-    op: detect
-    input_filename: index.html
-    expected_content_type: generic_html
+- id: "1.4"
+  generate: binary_noise
+  filename: archive.xyz
+  detect: unknown
 
-  - id: "1.3"
-    description: "Detect plain text"
-    score_name: preprocess.detect_text
-    file: notes.txt
-    op: detect
-    input_filename: notes.txt
-    expected_content_type: plain_text
+- id: "1.5"
+  file: email_action.html
+  process: email_html
+  no_html: true
+  min_chars: 50
+  ratio_lt: 0.8
 
-  - id: "1.4"
-    description: "Detect unknown (binary-like content)"
-    score_name: preprocess.detect_unknown
-    generate: binary_noise
-    op: detect
-    input_filename: archive.xyz
-    expected_content_type: unknown
+- id: "1.6"
+  file: email_action.html
+  process: email_html
+  has_meta: [subject, from]
 
-  # ── Preprocess tests ───────────────────────────────────────────────
+- id: "1.7"
+  file: email_thread.html
+  process: email_html
+  contains: "Sure, I'll handle the deploy"
+  excludes: "Let's plan the deploy"
 
-  - id: "1.5"
-    description: "Email: strip HTML tags"
-    file: email_action.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      no_html_tags: true
-      min_length: 50
-      compression_ratio_lt: 0.8
+- id: "1.8"
+  file: email_single.html
+  process: email_html
+  contains: "deploy is done"
 
-  - id: "1.6"
-    description: "Email: extract metadata (Subject + From)"
-    file: email_action.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      metadata_keys: [subject, from]
+- id: "1.9"
+  file: email_heavy.html
+  process: email_html
+  no_html: true
+  min_chars: 30
+  excludes: [border-collapse, font-size]
 
-  - id: "1.7"
-    description: "Email: split thread — solo ultimo messaggio"
-    file: email_thread.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      contains: "Sure, I'll handle the deploy"
-      not_contains: "Let's plan the deploy"
-
-  - id: "1.8"
-    description: "Email: singolo messaggio senza thread"
-    file: email_single.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      contains: "deploy is done"
-
-  - id: "1.9"
-    description: "Email: HTML pesante con table layout"
-    file: email_heavy.html
-    op: preprocess
-    input_content_type: email_html
-    assertions:
-      no_html_tags: true
-      min_length: 30
-      not_contains:
-        - "border-collapse"
-        - "font-size"
-
-  - id: "1.10"
-    description: "Fallback: file sconosciuto → testo restituito"
-    file: fallback.txt
-    op: preprocess
-    input_content_type: unknown
-    assertions:
-      min_length: 1
-      content_type: unknown
+- id: "1.10"
+  file: fallback.txt
+  process: unknown
+  min_chars: 1
+  content_type: unknown