feat: replace embedchain with native crewai adapter (#451)

- Remove embedchain adapter; add crewai rag adapter and update all search tools  
- Add loaders: pdf, youtube (video & channel), github, docs site, mysql, postgresql  
- Add configurable similarity threshold, limit params, and embedding_model support  
- Improve chromadb compatibility (sanitize metadata, convert columns, fix chunking)  
- Fix xml encoding, Python 3.10 issues, and youtube url spoofing  
- Update crewai dependency and instructions; refresh uv.lock  
- Update tests for new rag adapter and search params

src/crewai_tools/rag/chunkers/base_chunker.py

@@ -112,7 +112,10 @@ class RecursiveCharacterTextSplitter:
                 if separator == "":
                     doc = "".join(current_doc)
                 else:
-                    doc = separator.join(current_doc)
+                    if self._keep_separator and separator == " ":
+                        doc = "".join(current_doc)
+                    else:
+                        doc = separator.join(current_doc)
 
                 if doc:
                     docs.append(doc)
@@ -133,7 +136,10 @@ class RecursiveCharacterTextSplitter:
             if separator == "":
                 doc = "".join(current_doc)
             else:
-                doc = separator.join(current_doc)
+                if self._keep_separator and separator == " ":
+                    doc = "".join(current_doc)
+                else:
+                    doc = separator.join(current_doc)
 
             if doc:
                 docs.append(doc)