fix: build libnvidia-container without network access

[kvinwang]

Summary

• Add GOFLAGS=-mod=vendor to use the existing vendor directory in the source tree, eliminating Go module downloads during do_compile
• Fix nvidia-modprobe version mismatch: recipe fetched 550.40.07 via SRC_URI but the source Makefile expected 550.54.14 (the old do_compile[network]=1 masked this by re-downloading at build time)
• Apply nvidia-modprobe.patch in do_configure since we now skip the Makefile's curl-based download rule
• Remove do_compile[network] = "1"

Test plan

• bitbake libnvidia-container builds successfully without network in do_compile
• Full build.sh guest image build passes

Partially addresses #46 (libnvidia-container only; dstack-guest and dstack-sysbox are not feasible to fully offline, see comment on the issue)

[Copilot]

Pull request overview

Updates the Yocto libnvidia-container recipe to build reproducibly without requiring network access during do_compile, aligning with build isolation expectations in meta-nvidia.

Changes:

• Force Go builds to use vendored modules via GOFLAGS=-mod=vendor and remove do_compile[network] = "1".
• Fix nvidia-modprobe source/version mismatch by updating NVIDIA_MODPROBE_VERSION and the pinned SRCREV_modprobe.
• Apply mk/nvidia-modprobe.patch during do_configure and mark the dependency as downloaded to avoid the upstream curl-based download rule.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
meta-nvidia/recipes-graphics/libnvidia-container/libnvidia-container_1.00.bb	Enforces offline build behavior, applies the modprobe patch during configure, and removes networked compile.
meta-nvidia/recipes-graphics/libnvidia-container/libnvidia-container.inc	Aligns the fetched nvidia-modprobe revision with the version expected by the build.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

do_configure:append() applies nvidia-modprobe.patch unconditionally. If the task is re-run (e.g., -c configure -f, sstate invalidation, or iterative dev builds), patch will typically fail with “Reversed (or previously applied) patch detected”, causing a hard build failure. Make the patch step idempotent (e.g., gate on a dedicated stamp file, or use patch --forward/--batch and handle the already-applied case) and only create the stamp after a successful apply.

Suggested change

	cd ${S}/deps/src/nvidia-modprobe-${NVIDIA_MODPROBE_VERSION}
	patch -p1 < ${S}/mk/nvidia-modprobe.patch
	touch ${S}/deps/src/nvidia-modprobe-${NVIDIA_MODPROBE_VERSION}/.download_stamp
	modprobe_src="${S}/deps/src/nvidia-modprobe-${NVIDIA_MODPROBE_VERSION}"
	cd ${modprobe_src}
	if [ ! -e ${modprobe_src}/.download_stamp ]; then
	patch -p1 < ${S}/mk/nvidia-modprobe.patch
	touch ${modprobe_src}/.download_stamp
	fi