CLI export exits successfully with invalid archive for large SQLite site

## Problem

While validating draft PR #3997 against a real, large Studio site before SQLite-to-MySQL conversion, the supported full-site export command exited successfully with no diagnostic but produced an invalid archive.

## Reproduction

Integration build: current `origin/trunk` plus PR #3997, commit `3e547db03` in a managed test worktree.

```sh
node apps/cli/dist/cli/main.mjs export
/Users/.../intelligence-chubes4-full.zip
--path /Users/.../intelligence-chubes4
--mode full
```

Source SQLite database size: `19,880,640,512` bytes.
Produced archive size: `8,440` bytes.

Validation:

```text
End-of-central-directory signature not found
unzip: cannot find zipfile directory
```

The command returned success and printed no warning/error. The source site remained unchanged.

## Expected

A successful export produces a readable full-site archive. Any interrupted, unsupported, capacity-limited, or failed export exits non-zero, removes/marks the partial artifact, and reports the failing phase with actionable diagnostics.

## Why this matters for #3997

`studio convert --to mysql` is intentionally being validated first against a clone with a verified rollback artifact. Silent export corruption makes the documented safe migration path appear ready when no usable rollback exists.

## Acceptance criteria

- Large SQLite full exports produce a valid archive or fail non-zero.
- Partial archives are never reported as successful.
- Export performs a final archive-integrity check before success.
- Regression coverage includes an interrupted/failed archive stream and bounded large-file behavior.

## AI assistance
- **AI assistance:** Yes
- **Tool(s):** OpenAI GPT-5.6 Sol via OpenCode
- **Used for:** Integrating #3997 with current trunk, running the migration preflight, and capturing deterministic export evidence.

Start by reproducing the export command shown against a large SQLite site and trace the CLI entry point at apps/cli/dist/cli/main.mjs, including the archive-writing phase. Add regression coverage for interrupted or failed archive streams and bounded large-file behavior; done means valid archives succeed, failures exit non-zero with diagnostics, partial artifacts are not reported as successful, and final integrity is checked.

Written by the indexing model from the issue text.