fix(pixel-inference): validate finite float bounds for temperature parameter

[Vishaaallll]

Why this matters

In ods/extensions/services/pixel-inference/app/main.py, FIELDS accepted temperature in the request payload, but _prepare(payload, grant) omitted range and type validation for the parameter. If a caller supplied non-numeric types, boolean values, non-finite values (NaN, inf), or values outside the standard [0.0, 2.0] sampling range, the invalid value passed unvalidated to the backend inference runtime, causing runtime crashes or undefined sampling distribution behavior.

This surgical fix validates that temperature, when present, is a non-boolean finite float or integer within [0.0, 2.0], raising ShareError(400, 'invalid_temperature') otherwise.

Validation

• Baseline reproduction: verified that _prepare({"model": "ods/shared", "messages": [{"role": "user", "content": "hi"}], "temperature": -0.5}, grant) passed without error on unpatched code.
• Post-fix behavior: properly rejects negative, excessive (>2.0), non-finite, and non-numeric temperature values with HTTP 400 invalid_temperature.
• Telemetry: pixel-inference suite: 2 passed. New-test Ruff, syntax, and diff checks pass; new regression wired into Linux CI.

Overlap check

Searched open/closed PRs on Osmantic/ODS. PR #5616 touched capabilities limits and #5696 touched settings contracts. Neither addresses request-time inference parameter bounds in pixel-inference/app/main.py.

Risk / AI disclosure

AI-assisted investigation, patch, and regression tests. Strictly validates incoming HTTP inference payload parameters against the OpenAI specification. Independent review remains a gate.

Follow-up integration evidence

Composed locally with public-beta at 4fa6ad170 without conflicts. Regression tests pass locally and in Linux CI.