Add Function to Preprocess Request Prompt & Address String Escape Symbol Issue

[davidgxue]

Description

• See GraphQL Syntax Error Due to String Escape Character in LangSmith Log History #278 for details of why the error occurs. This won't be explained again in this PR.
• After investigation, it seems like LangChain's implementation, specifically how Langchain calls Weaviate's hybrid retriever is not optimal. Specifically, see this discussion thread from Weaviate https://forum.weaviate.io/t/hybrid-search-resulting-in-graphql-error/539. Technically, the safer way to build the query is using .build() instead of .do(). Using build() would have prevented symbols such as \ in the original string to be properly transformed into GraphQL syntax compatible string. However, LangChain implemented this part of the code using .do() which causes issues with certain symbols that conflict with the GraphQL syntax. See here https://github.com/langchain-ai/langchain/blob/master/libs/community/langchain_community/retrievers/weaviate_hybrid_search.py#L152 on line 152.
• As a result, we can only work around this issue by getting rid of the problem causing string escape symbol \. We can achieve this by parsing any single standalone \ used to do string escape out of the original query. This shouldn't affect the original meaning of the prompt as the \ gets auto added downstream anyway when querying the LLMs with string escape needed so that is not a concern.

Technical Changes

• Add a _preprocess_request function that does input validation and transformation on the user prompt to workaround this issue in LangChain.

Tests

1. When question exceeds character limit
   

2. When thread has more 6 questions/answers pairs, checked langsmith and saw only most recent 5 pairs

3. Questions with more than 1 \ is kept the same and questions with a single \ the \ is removed.
   For example, the original github issue's error causing query is now being successfully executed.
   Enter prompt: KeyError: \'gcp\' (note in Langsmith logs the prompt it auto adds another backslack to existing backslashs`)

closes #278

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Pages

Latest commit:	b166010
Status:	✅  Deploy successful!
Preview URL:	https://7484616d.ask-astro.pages.dev
Branch Preview URL:	https://add-request-preprocessor.ask-astro.pages.dev

View logs

[davidgxue]

Addressed your comment about custom error type in the new commit. I changed up the error propagation logic a little bit as well to make it cleaner. Thanks for the help reviewing @Lee-W @sunank200