feat: use original input for RAGTool / VectorDBTool

[SuZhou-Joe]

Description

In conversational agent, LLM is transforming user's input to other input like JSON / DSL query etc. So for VectorDBTool / RAGTool, we need to keep original input.

Issues Resolved

[List any issues this PR will resolve]

Check List

• New functionality includes testing.
  • All tests pass
• New functionality has been documented.
  • New functionality has javadoc added
• Commits are signed per the DCO using --signoff

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.
For more information on following Developer Certificate of Origin and signing off your commits, please check here.

[zane-neo]

This is a new method defined in Tool's interface?

[SuZhou-Joe]

Nope, it's an existing method but none of the tools override it. For RAGTool and VectorDBTool, they need original input(plain text) instead of a LLM-transformed input.

[zane-neo]

Can you add some simple comments to describe why original input is needed for this two tools?

[mingshl]

Thanks @SuZhou-Joe to help flipping on the flag following the fix in common opensearch-project/ml-commons#2022 using original input for RAGTool/VectorDBTool/NeuralSparseSearchTool. IMO, the neuralSparseSearchTool also need to take original input question instead of llm generated questions. @zhichao-aws would you agree on this for NeuralSparseSearchTool?

In this case, since vectorDB and NeuralSparseSearchTool are both extending from AbstractRetrieverTool, we can modify the AbstractRetrieverTool to flip on UseOriginalInput to true. Then both VectorDB and NeuralSparseSearchTool are using original input.

RAGTool also need to keep this useOrginalInput flag because it's not extending from AbstractRetrieverTool now.

[SuZhou-Joe]

Hi team, we are trying to do a JSON serialization when finding that the input question is a JSON syntax @zhichao-aws . The new solution has some pros:

• It can still leverage the LLM's output to increase search accuracy.
• It will address the issue that if we keep the original input and input happens to have" inside, the buildQuery method will throw error as it break the structure of a standard JSON.

[zhichao-aws]

The root cause is we format a query string (json string) in search tools. The query text is expected to be a plain text. If the query text itself is json string, it will throw exceptions to parse the formated string. Fix this in #203

[zhichao-aws]

#203 is merged now. Let's close this PR. We can re-open it on-demand.