shell/syntax: bound feature metadata lookahead

[ewhauser]

Summary

• bound parameter-expansion metadata lookahead so unsupported-feature errors stop after the bytes needed for classification
• preserve stable flag/index metadata across chunk boundaries without draining the rest of the reader
• add a regression proving the parser leaves later reader chunks unread when returning the first LangError

Why

• the merged parameter-expansion metadata fix solved chunk-boundary instability, but its first implementation used ReadFrom(p.src) and could block on long-lived readers while collecting metadata
• this follow-up keeps the metadata stable while avoiding whole-stream consumption on the error path

Validation

• make lint
• go test ./shell/syntax -run "TestParseLangErrorFeatureMetadata|TestParseLangErrorFeatureMetadataDoesNotDrainReader" -count=1
• go test ./shell/syntax -count=1

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 229b14e393

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".