fix(model-contract): coerce integer float token limits in projection

[vaibhavsrv]

Why this matters

In ods/bin/pixel_model_contract.py, the model limits resolver projection() evaluates override parameters across SDK defaults, selected model params, and agent configs. When parsing maxTokens, max_completion_tokens, or max_tokens, line 78 explicitly admits finite positive floating-point values: if (type(value) is int or type(value) is float and math.isfinite(value)) and value >= 0: output, resolved = value, True. However, line 84 binds maxOutputTokens: output, and downstream boundary assertions (lines 87 and 89) strictly mandate type(limits[key]) is not int to raise ModelError("invalid-model-limits"). Consequently, configurations containing integer-valued floats (e.g. 4096.0, often produced by JSON/YAML parsers, SDK float serialization, or calculated budgets) are admitted at resolution time but immediately fail contract projection with ModelError('invalid-model-limits').

This fix surgically checks value.is_integer() and casts output = int(value) when resolving token limit overrides in projection(). Fractional floats (such as 4096.5) or non-finite values are rejected, preserving discrete token count constraints while ensuring integer-valued floats cleanly project to valid integers. Existing contract validation, compaction defaults, and route binding semantics remain untouched.

Validation

• Baseline reproduction: Passing {"params": {"max_tokens": 4096.0}} into projection(config) admitted the value into output but subsequently crashed at the type(limits[key]) is int check with ModelError('invalid-model-limits').
• Post-fix behavior: Integer-valued floats (4096.0) resolve cleanly to integer 4096 in limits["maxOutputTokens"], while non-integral floats (4096.5) and non-finite floats continue to be rejected with ModelError('invalid-model-limits').
• Telemetry statement: "[Model-contract] suites: 4 passed. New-test Ruff, ShellCheck, and diff checks pass; new regressions wired into Linux CI."

Overlap check

• PR fix(pixel-relay): return 502 bad gateway when upstream model router is unavailable #5605 (fix(pixel-relay): return 502 bad gateway when upstream model router is unavailable): Addressed reverse proxy upstream connectivity in pixel-model-relay; does not touch model contract metadata projection in ods/bin/pixel_model_contract.py.
• PR fix(pixel-edge): accumulate content before finish reason in stream fallback guard #5604 (fix(pixel-edge): accumulate content before finish reason in stream fallback guard): Focused on edge runtime streaming chunks; no overlap with model contract limits validation.
• PR fix(pixel): guard task group exception groups during stream revocation cleanup #5600 (fix(pixel): guard task group exception groups during stream revocation cleanup): Addressed task group cleanup; no overlap with contract resolution.

Risk / AI disclosure

AI-assisted investigation, implementation and CLI regressions. This strengthens model contract token limits resolution, not runtime admission. Independent human review and platform/runtime qualification remain gates. No running configuration, deployment or upstream merge changed.

Follow-up integration evidence

Composed with #5605 at ef21605 without conflicts. Production and test diffs passed together; pixel model transition and contract checks remain intact.
Backlog composition was local-only (production/test diffs, excluding workflow/Makefile wiring); it is not an upstream merge or independent human approval. Declared live-review gates remain open.