langchain4j-ingest: document parsers for binary formats (tika, docling)

Ingestion pipelines read every payload as plain text, so the documents people actually want to ingest — PDFs, office files, HTML — come out as garbage instead of their content.

A pipeline should be able to name a parser that extracts the text first:

```properties
quarkus.camel.langchain4j.ingest.manuals.parser=tika
```

Two options, reusing the existing Camel components: `tika` extracts text in-process, `docling` converts to markdown through a Docling Serve instance. Works for both directory and consumer-fed pipelines, in JVM and native mode.

Start by locating the langchain4j-ingest pipeline configuration and the existing Camel Tika and Docling components. Trace parser selection for both directory and consumer-fed pipelines, then check JVM and native-mode coverage. Done means binary documents yield extracted text or markdown through the configured parser in all four combinations.

Written by the indexing model from the issue text.