chore: reverts judge stripping behavior

[andrewklatzke]

Requirements

• I have added test coverage for new or changed functionality
• I have followed the repository's pull request submission guidelines
• I have validated my changes against all supported platform versions

Related issues

https://launchdarkly.atlassian.net/jira/software/c/projects/AIC/boards/2045?selectedIssue=AIC-2848

Describe the solution you've provided

This is a reversion of functionality that was added.

Describe alternatives you've considered

Reverts behavior from #1364 that stripped messages off of the judges and then re-created them.

Additional context

This was discussed and there are various concerns about this behavior. Since this is a semi-permanent solution once changes are applied on the server side we don't want to take this step yet.

---

Note

Medium Risk
Judge configs again expose full legacy message lists to callers, which can change runtime prompts versus the stripped behavior. The widened Runner.run signature is a public API change, though string-based call paths stay backward compatible.

Overview
Reverts judge message stripping from a prior change: stripLegacyJudgeMessages, reserved-variable re-injection for message_history / response_to_evaluate, and related warnings are removed. judgeConfig now returns flag-evaluated messages unchanged (including legacy {{message_history}} / {{response_to_evaluate}} template rows).

Extends the Runner protocol so run accepts string | LDMessage[]. String inputs keep the existing behavior (config messages prepended, multi-turn history updated). A pre-built LDMessage[] is sent to the provider as-is—no config merge and no history mutation—implemented in the OpenAI, LangChain, and Vercel model runners with matching tests.

^{Reviewed by Cursor Bugbot for commit 1d6223b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

@launchdarkly/js-sdk-common size report
This is the brotli compressed size of the ESM build.
Compressed size: 26365 bytes
Compressed size limit: 29000
Uncompressed size: 129044 bytes

[github-actions]

@launchdarkly/js-client-sdk-common size report
This is the brotli compressed size of the ESM build.
Compressed size: 38739 bytes
Compressed size limit: 39000
Uncompressed size: 212244 bytes

[github-actions]

@launchdarkly/browser size report
This is the brotli compressed size of the ESM build.
Compressed size: 179579 bytes
Compressed size limit: 200000
Uncompressed size: 831422 bytes

[github-actions]

@launchdarkly/js-client-sdk size report
This is the brotli compressed size of the ESM build.
Compressed size: 31978 bytes
Compressed size limit: 34000
Uncompressed size: 114243 bytes

[joker23]

Should the agent runners also be updated (eg

js-core/packages/ai-providers/server-ai-openai/src/OpenAIAgentRunner.ts

Line 31 in a5b42ca

export class OpenAIAgentRunner implements Runner {

) now that the base runner interface can take in string | LDMessage[]?

[andrewklatzke]

Should the agent runners also be updated (eg

js-core/packages/ai-providers/server-ai-openai/src/OpenAIAgentRunner.ts

Line 31 in a5b42ca

export class OpenAIAgentRunner implements Runner {

) now that the base runner interface can take in string | LDMessage[]?

I don't believe so. Agent runners should only be taking in a single instructions block, not an array of messages.

Waiting on @jsonbailey for reviews on these though since it's reverting behavior