ci(native): add force input to rebuild a drifted artifact key

[lxm7]

Why

The iOS EAS build has been failing in eas-build-pre-install, before any compile:

native(ios): CHECKSUM MISMATCH — chat_mls_coreFFI-ios-7e4fdcc9abc3f190.tar.gz does not match committed chat_mls_coreFFI.xcframework.sha256

The key is correct and the fetch succeeds — the published asset just doesn't match the committed manifest. In the tarball, headers/plists/modulemaps verify OK; only the two Mach-O binaries and the top-level Info.plist fail. Same source, different compile.

What happened: CI run 28983282244 built, uploaded, and committed the manifest in b8bf0fa on Jul 8 23:54. The R2 object's Last-Modified is Jul 9 09:50 — ten hours later, with no workflow run in between. A locally built xcframework was uploaded over the CI-published one (confirmed: the local ios-arm64 binary hashes 2a2f899c76e8dc…, byte-identical to what R2 serves, and both fail the committed manifest).

Because the binaries are non-reproducible, the same source key yields different bytes per compile — so the consumer can never verify, and every iOS EAS build fails until the two sides are realigned.

The trap

Already published? only checks existence (curl -fsI). Once a key drifts, it's unrepairable by dispatch: the guard sees the object, skips the build, and nothing republishes. The only escape was deleting the R2 object by hand.

What this does

Adds a force boolean to workflow_dispatch that bypasses the guard on both jobs. aws s3 cp overwrites in place, and the commit step refreshes the manifest back into lockstep with the asset.

inputs.force is null on push/PR events, so those paths are unchanged. Passed via env: rather than interpolated into run:.

Follow-up (not in this PR)

Two things would stop this recurring:

1. Make R2 writes non-clobbering — aws s3api put-object --if-none-match "*" so an upload against an existing key fails loudly instead of poisoning it.
2. Make the guard verify, not just exist — fetch and check against the committed manifest, so drift surfaces in CI rather than in an EAS hook hours later.

Local build:ios rewriting the committed manifest is the mechanism the drift travels through; worth a guard there too.