fix: correct mistralai import path — Mistral is in mistralai.client

feat: switch entire backend from Gemini to Mistral (LLM + STT)

feat: instant LinkedIn PDF parsing via heuristics, LLM as fallback

perf: reduce PDF parser token budget 16K → 4K for faster parsing

fix: add 60s timeout to Gemini call in PDF parser

fix: increase max_output_tokens to 16000 — Gemini 2.5 Flash thinking was consuming output budget, truncating JSON

fix: robust JSON extraction from Gemini 2.5 Flash — find JSON in response, strip thinking, fix trailing commas

fix: use gemini-2.0-flash-lite for PDF parsing — no thinking interference, reliable JSON

fix: response_mime_type=application/json on ALL Gemini calls — no more broken JSON, null handling, guaranteed valid output

fix: remove thinking_config (breaks SDK), restore working PDF parser

perf: disable thinking for PDF parsing + drafts, background analysis, instant chat navigation

feat: full CV preview — personal info, education, languages, structured sections with labels

feat: LLM-powered PDF parser (handles real LinkedIn format), focused first question with company names

feat: LinkedIn PDF parser — extract profile, experiences, education, skills