iOS pipeline: self-recovering simulator builds + docs

[shoeless]

Problem

RoboVM 2.3.24's robovm:ipad-sim mojo launches the app after the AOT link and then blocks in SimLauncherProcess.waitFor until the app quits — and on Apple Silicon it picks its own simulator (its device-type table predates arm64 sims). Net effect: every scripted simulator build hung at the launch step and needed the wedged maven process killed by hand.

Fix

All the pipeline needs from the mojo is target/robovm.tmp/config.xml, written once the AOT link finishes. So sim mode now:

1. runs the mojo in its own process group with output to a log,
2. waits for config.xml to appear,
3. stops the process group right before the launch step it would wedge in,
4. rebundles with the standalone AppCompiler via the existing assemble_arm64_sim_app (the AOT cache is content-hashed — nothing recompiles).

Also adds pipeline/README.md documenting the transform stages (JvmDowngrader → MobiVmBridge → MobiVmLinkAudit) and the CI audit mode, plus comment-only clarifications in the two pipeline helpers.

Verification

Ran pipeline/ios-pipeline.sh classpath && pipeline/ios-pipeline.sh sim on this branch against current master: the build completed build → install → launch on the simulator fully unattended — previously the same invocation required manually detecting and killing the wedged mojo on every single build.

Build-tooling only; no app/runtime code touched.

🤖 Generated with Claude Code

[tool4ever]

By the way, I assume the code block in "ios/Main.java" related to cardcache.bin is obsolete?

[shoeless]

Trimmed to just the functional change — the README and comment churn are gone (the doc you moved to docs/Development/iOS-Builds.md covers it). Two corrections from re-testing: the early-stop is now gated to Apple Silicon — on Intel the mojo bundles the .app during its launch phase, so stopping at config.xml would have broken those builds — and the wait tests -s with a settle sleep, since the mojo opens config.xml before serializing into it.

And correct about cardcache.bin — nothing in the repo writes those files (they're left over from a card-cache experiment on my iOS branch); dropped the block in this PR.