Link the iOS archive without ForceLoad

[jcant0n]

The first run that reached the linker died on duplicate symbols:

duplicate symbol '_WebPInitUpsamplersNEON'
duplicate symbol '_jpeg_destroy_decompress'
duplicate symbol '_jsimd_ycc_extrgb_convert_neon'

all of them twice inside libCesiumNativeC.a itself. The merge in CesiumC takes every .a it finds, and two pairs among them overlap by design: libjpeg.a and libturbojpeg.a are two APIs over one core, and libwebpdecoder.a is a subset of libwebp.a.

ForceLoad is what turns that into an error — it pulls every member into the link whether or not anything references it, so both copies arrive. Ordinary linking takes the first definition and moves on, which is what browser-wasm has always done with the same archive and the same overlap.

What ForceLoad was protecting against does not go away

The linker keeps only members something references, and a symbol reached solely through a P/Invoke can be dropped. The smoke leg is what makes turning this off safe rather than hopeful: it reads the built executable and fails if cesium_async_system_create is not among its defined symbols — exactly the failure this could cause.

That check was written before this problem existed, for a different reason. It is the reason this can be a one-line change rather than a guess.

Good news in the same run

The macos-latest move worked: Xcode 26.6, and the build got all the way to clang++. So the targets file is putting the archive on the link line — the half of phase 4 that had never been exercised.

The duplication itself is worth removing in CesiumC, since it also makes a 61 MB archive bigger than it needs to be and any consumer passing -force_load hits the same wall. Separate change, needs a release cycle, tracked.

[github-actions]

API gate: additive

Every symbol that existed still exists, unchanged. Nothing that compiled before stops compiling.

Target framework	Added	Removed
net10.0-ios	+0	-0
net10.0	+0	-0

_{Enum and constant values are part of the measured surface: a renumbering keeps compiling and sends the wrong number to the driver, so it counts as a removal.}