main does not compile under MSVC, and main-baseline.py reports GREEN because windows-msvc-* are PR-only jobs it never runs

[localai-bot]

main is red on two gates its own baseline lane never runs

windows-msvc-cpu and windows-msvc-vulkan are gated
if: github.event_name == 'pull_request'
(ci.yml:640), so the scheduled
and dispatched lane that scripts/main-baseline.py reads never executes
them. main-baseline.py therefore reports:

NEWEST BASELINE: GREEN at bbc482a2de73ae0c433f210c5780b52d6c886f5d

while both jobs fail on every pull request that reaches them. The baseline is
answering a question about a different set of jobs than the one a PR author is
looking at, and nothing says so.

The defect itself

tests/vt/test_cpu_isa_x86.cpp does not include <ostream>. On libstdc++ it
arrives transitively; MSVC's <string_view> does not provide it, so doctest's
stringification instantiates operator<< against an undefined
std::basic_ostream:

__msvc_string_view.hpp(550,23): error C2027: use of undefined type
    'std::basic_ostream<char,std::char_traits<char>>'
    [build-pr-windows-vulkan\tests\test_cpu_isa_x86.vcxproj]
__msvc_string_view.hpp(550,23): error C2065: 'iostate': undeclared identifier
...
build-windows-release.ps1:25  throw "$Program exited with status $LASTEXITCODE"
    cmake exited with status 1

The file's includes are <array> <bit> <limits> <stdexcept> <string>, then
doctest/doctest.h. No <ostream>, and no <iostream>.

Reproduced on two unrelated branches, byte-for-byte

branch	head	job	first error
row/ENG-RELEASE-WINDOWS	673c2f3d	windows-msvc-vulkan	__msvc_string_view.hpp(550,23): error C2027
row/BENCHMARKS-CAP-RELIEF (#494)	104d3f36	windows-msvc-vulkan	__msvc_string_view.hpp(550,23): error C2027

Same file, same line, same failing target, same failing step ("Build and execute
the native Windows Vulkan focused gate"). #494 is a Python-only change to two
governance checkers and their mutation suites: it touches nothing under src/,
include/, tests/, cmake/, or any .ps1, so it cannot be the cause. Every
PR that reaches these two lanes gets a red X it did not earn.

Why it went unnoticed

tests/vt/test_cpu_isa_x86.cpp was last touched by #117 (49539559,
8fa2ecdb, 94f72e14), the Windows release row. The lane that would catch a
Windows-only include defect is the lane that only runs on PRs, and the row that
introduced it is itself the row whose PRs are red, so the signal reads as "the
Windows row is still in progress" rather than "main does not compile under
MSVC".

Repair

Two separable things, and the second is the one that matters:

1. Add the missing <ostream> to tests/vt/test_cpu_isa_x86.cpp. RED-before is
   the MSVC build; green-after is the same lane on a PR.
2. Make the main baseline honest about jobs it does not run.
   scripts/main-baseline.py already prints a not run: list (pr-size,
   commit-protocol-tag, documentation-checkpoint). windows-msvc-cpu and
   windows-msvc-vulkan are not in it, because they are not defined for that
   event at all, so they are invisible rather than listed. A GREEN verdict
   that silently excludes two compiling gates is the same failure mode as a
   green test run with cases skipped.

Per AGENTS.md, (2) changes a checker's semantics and needs its own spec,
red-before and green-after, with the mutation in
tests/scripts/test_main_baseline.py.

Not fixed in #494, deliberately

Found while doing #460. Not fixed there because I cannot produce green-after
evidence for it: there is no MSVC toolchain on this box, the lane is
Windows-only, and a one-line include change I cannot compile is a guess, not a
fix. tests/vt/test_cpu_isa_x86.cpp also belongs to row/ENG-RELEASE-WINDOWS,
which is in flight and red on the same job, so patching it from an unrelated row
would collide with the branch already working there.

[localai-bot]

Another instance of this class found while landing #699 (a records-only PR): src/vllm/multimodal/video_engine.cpp has three unguarded POSIX sites, filed as #729. Same shape as #603 — landed green on main because these jobs are PR-only, then reds every PR after it.