fix: complete tabular outputs and public workspace chat grounding

[Paul Lizer (paullizer)]

Summary

• recognizes row/line terminology and activates durable tabular routing for upgraded deployments by migrating stale backend-only settings
• preserves the selected model endpoint through Analyze preflight, hardens artifact-set publication/recovery, and returns safe failure details
• makes explicit per-row narrative requests exhaustive: Search returns one all-row Markdown artifact, while Analyze returns a concise Markdown summary plus an all-row Markdown sibling
• preserves explicit structured behavior: Search CSV/JSON/XML returns the requested artifact; Analyze returns Markdown plus requested structured siblings
• restores grounded chat for explicitly selected documents in accessible public workspaces hidden from the public directory

Tabular behavior matrix

Request	Search	Analyze
Explicit CSV/JSON/XML	Requested structured artifact	Markdown summary plus requested structured artifact
Explicit per-row/per-line narrative, no format	One exhaustive row-by-row Markdown artifact	Markdown summary plus exhaustive row-by-row Markdown artifact
Aggregate all-row themes/risks	Bounded hierarchical Markdown summary	Bounded hierarchical Markdown summary

Exact-row Markdown persists ordered questions as answer_1 through answer_N, uses one checkpointed result per source row, requires every answer to be non-empty, validates source order and final row count, and escapes untrusted data as literal Markdown.

Validation

• Python compilation and JavaScript syntax checks pass
• deterministic 200-row × 8-question test validates 200 row sections and 1,600 answers, including the final row/final answer
• malformed answer sequences, missing answers, question/schema mismatch, active Markdown links, and raw HTML are rejected or escaped
• Search/Analyze planner, preflight, artifact lifecycle, public status, document-action, and production-correctness suites pass
• full tabular scale suite passes 70/70, including 100,000-row planning, retries, leases, resumability, and idempotent publication
• full background output Playwright suite passes 7 tests with 4 environment-dependent skips
• git diff --check and untracked EOF/whitespace checks pass
• independent final review found no critical or high-severity blockers

Fixes #1233
Fixes #1245