Add LangChain4j document ingestion Kamelets

The catalog has no Kamelets for the ingestion half of a RAG setup — feeding documents into a vector store. Apache Camel is gaining a `camel-langchain4j-ingest` component (CAMEL-24621, apache/camel#26190, targeted at Camel 4.23) that provides the engine: split, batch-embed and store into a LangChain4j `EmbeddingStore`, with identity metadata on every segment and optional deduplication by document id.

This issue proposes four Kamelets on top of it:

- **langchain4j-ingest-sink** — the ingestion step; any source can feed it
- **tika-extract-text-action** / **docling-convert-action** — generic document→text/markdown parse steps (useful beyond ingestion), capturing the document id before the parse so a crafted document cannot forge its own identity
- **langchain4j-ingest-file-source** — a directory source with knowledge-base semantics: files are read, not consumed; edited files are re-delivered; parser actions receive the raw bytes

Citrus tests cover each Kamelet and the source → action → sink compositions, including a binary PDF. The tests require Camel 4.23+ and fail on older versions by design — the red build gates the merge until the catalog moves to a Camel that ships the component.

_This issue was drafted by Claude Code on behalf of @JiriOndrusek._