Export a research job's report as a PDF.

Builds the same body the UI renders. The set of check_results considered depends on the job's agent:

• Default research jobs: grep.open_research, grep.research_follow_up, grep.report_revision.
• grep-brain parent jobs: also pulls sub-jobs' check_results (grep.expert_chat, grep.research_fork_chat) plus the scheduled/recurring/user-edit revision command_ids.
• Brain sub-jobs (grep-expert-chat-v1 / grep-research-fork-chat): same expanded set scoped to the sub-job itself.

From those rows, _select_primary_and_followups picks one primary report (mirroring ResearchJob.tsx — latest completed revision wins, falling back to a running revision with body, then open_research, then the first non-followup) and a list of appended sections (grep.research_follow_up and non-revision brain sub-reports). The primary's body comes from payload.report (preferred) or answer. Browserbase renders the resulting markdown via build_pdf_html → generate_pdf_from_html, with a basic markdown_pdf fallback if Browserbase is unavailable.