ci: cover 3.14t in the floors leg

[lesnik512]

Follow-up to #63, which landed the floors job. One line: the leg's matrix gains 3.14t.

Why

standard.md §7 asks for the floors job on every entry of the same matrix pytest uses, and that is not incidental wording — the section spells out the reason:

wheel coverage is per interpreter, so two interpreters that resolve the same versions can still disagree on whether those versions install

pytest here runs six entries; the floors leg runs five. The one it omits is 3.14t, which is the entry most likely to be the odd one out: a free-threaded build is where a floor's cp314t wheel can be missing while cp314 is fine and every other entry stays green. Covering five of six leaves that gap open silently — the leg passes and says nothing.

Verification

In a clean worktree off main, uv.lock deleted before each resolve (a stale lock makes uv sync skip re-resolution and gives a false pass):

uv python pin 3.14t
uv sync --all-extras --no-install-project --resolution lowest-direct --no-build
uv run --no-sync pytest .

Resolves wheel-only and 20 passed, on a genuinely free-threaded interpreter (sys._is_gil_enabled() is False). Same for 3.13 and 3.14. So this entry is green today; it is added to keep it that way. Local runs are macOS/arm64 — CI is the verdict on manylinux coverage.

What this deliberately leaves alone

The job is named lowest; §7's job table names it floors. That is a rename across the org, not this repo's to decide unilaterally, and it is being tracked separately — left out here so this stays a one-line change.

On #62

Closing #62 in favour of this. Its diff also bounded pydantic in [project.dependencies], and that was wrong: the failure it was reacting to came from the lint group's unbounded typing-extensions, not from the shipped floor. uv runs one resolution across every group, so a low typing-extensions holds pydantic back to a build whose pydantic-core has no wheel for the interpreter under test. #63 fixed it at the right layer, in the group where the unbounded name actually lives.

Checked directly — a clean 3.13, 3.14 and 3.14t venv each:

uv pip install --no-build --resolution lowest-direct 'faststream>=0.7,<0.8' 'modern-di>=3,<4'

installs on all three, resolving faststream==0.7.0 with pydantic==2.13.5. The shipped floor was never the problem, and #62 would have constrained every user's pydantic to buy a CI-only artifact.