fix: keep remote GPU providers selectable across host platforms

[atomantic]

Summary

• Stop evaluating remote GPU providers and their models against the PortOS caller's hardware. A fleet API or OpenCode TUI provider now stays selectable when the caller has a different OS or no NVIDIA GPU.
• Preserve hardware requirements and report remote compatibility as unknown; keep existing hardware checks for local endpoints and unset endpoints.

Test plan

• Passed server systemCapabilities tests (9 tests), including remote API/TUI and local endpoint regression coverage.
• Completed read-only Claude review; no issues reported, but additional prose made its strict verdict optional-inconclusive.