ci: restore Intel macOS (x86_64) build that was dropped in cea0029

[slackarea]

Follow-up to #416 (merged).

What happened

• ci: add Intel macOS (x86_64) build target #416 added the Intel macOS build (native x86_64 on the macos-13 runner).
• You then improved it in 4f117d6 ("ci: add intel macos pdfium binary") — committing libpdfium-x86_64.dylib, a checksum step, and a swap step so the Intel .app links the correct PDFium.
• Commit cea0029 ("feat: add chat agent routing") then removed both the macos-13 matrix entry and the Intel PDFium swap step. The subject is about chat routing, unrelated to CI, so this looks unintentional — a stray hunk that rode along with the feature commit.

Effect: no Intel macOS artifact has shipped since v0.5.0 (v0.5.0 / v0.5.1 / v0.5.2 releases have only aarch64.dmg). Intel Mac users are back to having no prebuilt binary.

This PR simply restores those two pieces — identical to your own 4f117d6 config:

• the macos-13 / x86_64-apple-darwin matrix row, and
• the Use x86_64 pdfium binary (macOS Intel only) step.

The committed libpdfium-x86_64.dylib is still in the tree, so no new binaries are needed. Native build, no cross-compile.

I verified v0.4.24–v0.5.2 all build to a working x86_64 .app/.dmg on a real Intel iMac. If the removal was in fact deliberate (e.g. runner cost or macos-13 retirement timeline), feel free to close — just flagging it in case it was accidental.

🤖 Generated with Claude Code

[slackarea]

Important follow-up — I think I now see why this was removed, and it may have been justified rather than accidental.

I re-ran the Intel build on a fork and the macos-13 job never gets a runner — it sits in queued indefinitely. Checking this repo's own history, the v0.4.26 run (28008961148) shows the same thing: the build (macos-13, x86_64-apple-darwin) job was queued from 2026-06-23 07:12 and cancelled ~23h later, having never started. GitHub appears to have retired/brown-ed-out the macos-13 hosted runner, so the label is accepted but never scheduled.

So restoring the matrix entry as-is would just re-introduce a job that hangs for ~24h and then fails. Given that, and that arm64→x86_64 cross-compile is blocked by lance-linalg (AVX-512, per 9f05d60), there may be no clean GitHub-hosted path to an Intel build right now.

Feel free to close this PR — I no longer think the removal was a mistake. Flagging the diagnosis in case it's useful. For our own Intel users we'll publish unofficial builds from a self-hosted Intel runner on our side.

[nashsu]

@slackarea thanks for your understanding, It's true that Github Runners no longer provide Mac Intel machines, I'll source some Intel Mac machines later and host myself.