feat(llm): add robust regex fallback parser for JSON responses (fixes #318)

[Diwakar-odds]

Program

• ECSoC 2026
• ELUSOC 2026
• Regular contribution (not part of a program)

What does this PR do?

Introduces a robust fallback utility (parse_llm_json) in backend/features/llm_analysis/parser.py that gracefully handles malformed JSON responses from LLM models. The utility uses regex to strip markdown code blocks and conversational filler, ensuring valid JSON extraction even if the LLM output is wrapped or syntactically noisy.

Related issue

Closes #318

Checklist

• I have added the correct program label (ECSoC26 or ELUSOC) to this PR
• I tested this change locally
• This PR addresses exactly one issue/feature (not a bundle of unrelated changes)
• I did not open this PR purely to farm contribution counts

[vercel]

@Diwakar-odds is attempting to deploy a commit to the Eshaan's projects Team on Vercel.

A member of the Team first needs to authorize it.

[Diwakar-odds]

Hi @eshaanag 👋, I have submitted the PR to fix issue #318.

PR Analysis & ECSoC26 Level Justification

This PR introduces a robust regex-based fallback parser (\parse_llm_json) into the LLM analysis logic, which handles noisy or malformed JSON responses returned by LLM endpoints (e.g., Markdown wrapping or conversational filler).

• Core Backend: It enhances the resilience of the data pipeline against LLM format hallucinations.
• Reliability/Resilience: This directly resolves parser crashes by falling back on regex structures (using
  e.search\ over nested braces/markdown syntax) when \json.loads\ strictly fails.
• Tests Added: Comprehensive unit tests covering 5+ parsing edge cases (bare JSON, Markdown wrapped, conversational filler, and structural errors) have been added to the test suite.

Given that this PR involves hardening a core backend component (LLM parsing resilience) against format instability, it qualifies as an L2/L3 (Core Backend Resilience) contribution.

Could you please review, and assign the \ECSoC26\ label (if missing) along with \good-backend\ or \Level 3? Thank you!