fix(server): portable exit-code on subprocess-alive=false path (windows build break)

[marksverdhei]

Why

Every Windows build job on PRs #62 and #63 is failing with:

tools/server/server-models.cpp:923:85: error: no member named 'return_status' in 'subprocess_s'

subprocess_s::return_status is defined only in the POSIX branch of vendor/sheredom/subprocess.h:402-418. The Windows variant uses hProcess instead, so the direct field access doesn't compile.

Fix

Replace the direct field access with subprocess_join(child_proc.get(), &child_exit) — same pattern already used at line 908 of the same file. subprocess_alive() has already reaped the zombie on POSIX, so subprocess_join() returns immediately with the cached status (no risk of re-blocking on a stuck child).

- this->update_status(name, SERVER_MODEL_STATUS_UNLOADED, child_proc->return_status);
+ int child_exit = 0;
+ subprocess_join(child_proc.get(), &child_exit);
+ this->update_status(name, SERVER_MODEL_STATUS_UNLOADED, child_exit);

Impact

• Unblocks Windows builds on PR feat(dflash): integrate DFlash block-diffusion speculative decoder (rebased on post-rewrite ht) #62 (DFlash rebase) and PR fix(tests): TBQ block-size + tolerances after 128-block migration #63 (TBQ test fix).
• No behavior change on POSIX — subprocess_join just returns process->return_status.
• Local CPU build passes.

Verified

• ✅ cmake -B build -DGGML_CPU=ON -DLLAMA_BUILD_APP=ON && cmake --build build --target llama-server succeeds locally.
• ✅ Only one file touched, 1 LOC of real change.