[CP-stable]Move iOS compile benchmarks to Mac_benchmark and pin Mac_benchmark to M1 - #193581

… M1 (flutter#192472)

Part of flutter#186414

This batch moves the 4 iOS compile benchmarks off physical-device bots
and onto the pinned benchmark pool:
* `basic_material_app_ios__compile`
* `flutter_gallery_ios__compile`
* `hello_world_ios__compile`
* `imitation_game_flutter` (task: `imitation_game_flutter__compile`)

These tasks only run `flutter build ios` and never launch an app on the
attached iPhone (details in [flutter#186414
(comment)](flutter#186414 (comment))).
On `Mac_arm64_ios` they tie up physical devices and, because that pool
mixes M1 and M4 Pro hosts, every metric is split into two disconnected
Skia Perf series (`device_type=iPhone_11` vs `iPhone_16_Pro`) with a
~40–60% timing gap.

Per [flutter#186414
(comment)](flutter#186414 (comment)),
instead of moving them to the unpinned `Mac_arm64` pool:

**Changes**
- `platform_properties.mac_benchmark`: `mac_model` `"Macmini8,1"`
(Intel, being turned down per
[flutter.dev/go/macos-intel-deprecation](https://flutter.dev/go/macos-intel-deprecation))
→ `"Macmini9,1"` (Apple Silicon M1 Mac mini), plus `cpu: arm64` and an
`arm64` tag.
- Moved the 4 targets from `Mac_arm64_ios` to `Mac_benchmark` with
`bringup: true`. They inherit `device_type: none` and `["devicelab",
"hostonly", "mac", "arm64"]` from the platform; dropping the `ios` tag
also stops the devicelab recipe from running `ideviceinfo` (which would
fail on a host-only bot). `flutter_gallery_ios__compile` keeps the
`ruby` dependency for CocoaPods, matching
`flutter_gallery_macos__compile`.
- `imitation_game_swiftui` stays on `Mac_arm64_ios` — it runs
`xcodebuild test` against a physical device.

**Heads-up for the metrics sheriff**
- All 13 existing `Mac_benchmark` macOS targets (`*_macos__compile`,
`*_macos__start_up`, `complex_layout_scroll_perf_macos*`,
`new_gallery_macos_impeller__transition_perf`,
`flutter_tool_startup__macos`, …) move from Intel to M1 hardware with
this change, so their baselines will shift (and the `arch` benchmark tag
changes from `intel` to `arm`). Expect one-time step changes on Skia
Perf rather than regressions.
- Host-only `Macmini9,1` capacity in `luci.flutter.prod` is ~56 bots (vs
~37 Intel `Macmini8,1`), shared with `Mac_arm64`; today that pool is all
M1, so the pin mostly future-proofs against M4 Pros being added.