feat: linha do tempo processual dos autos vinculados Pet16704

Substitui o gitGraph do Mermaid (ilegível com 3.437 commits) por uma
timeline interativa com eixo temporal real, filtros por instituição,
tipo de documento e pessoas mencionadas (extraidas via NER local com
pierreguillou/ner-bert-base-cased-pt-lenerbr).

pipeline/run_ner.py

@@ -0,0 +1,54 @@
+import json, re, time
+from collections import Counter, defaultdict
+from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline
+
+docs = json.load(open("docs.json"))
+texts = [d["text"] for d in docs]
+print(f"{len(texts)} documents to process", flush=True)
+
+name = "pierreguillou/ner-bert-base-cased-pt-lenerbr"
+tok = AutoTokenizer.from_pretrained(name)
+model = AutoModelForTokenClassification.from_pretrained(name)
+nlp = pipeline("ner", model=model, tokenizer=tok, aggregation_strategy="simple", device=-1)
+
+def merge_entities(ents, group="PESSOA"):
+    ents = [e for e in ents if e["entity_group"] == group]
+    ents.sort(key=lambda e: e["start"])
+    merged = []
+    for e in ents:
+        if merged and e["start"] - merged[-1]["end"] <= 1:
+            merged[-1]["end"] = e["end"]
+        else:
+            merged.append({"start": e["start"], "end": e["end"]})
+    return merged
+
+t0 = time.time()
+BATCH = 32
+all_people_per_doc = []  # list aligned with docs: list of names
+for i in range(0, len(texts), BATCH):
+    batch = texts[i:i+BATCH]
+    results = nlp(batch)
+    for text, ents in zip(batch, results):
+        merged = merge_entities(ents, "PESSOA")
+        names = [text[m["start"]:m["end"]].strip() for m in merged]
+        names = [n for n in names if len(n) >= 3]
+        all_people_per_doc.append(names)
+    if (i//BATCH) % 10 == 0:
+        elapsed = time.time()-t0
+        print(f"{i+len(batch)}/{len(texts)} docs, {elapsed:.1f}s elapsed", flush=True)
+
+print("total time", time.time()-t0, "s")
+
+# save raw per-doc results
+out = []
+for d, names in zip(docs, all_people_per_doc):
+    out.append({"id": d["id"], "names": names})
+json.dump(out, open("ner_people_per_doc.json","w",encoding="utf-8"), ensure_ascii=False)
+
+counter = Counter()
+for names in all_people_per_doc:
+    for n in names:
+        counter[n] += 1
+print("distinct raw person strings:", len(counter))
+for n,c in counter.most_common(50):
+    print(c, n)