fix: bound non-finite shell output limits

[hsusul]

Summary

Malformed shell tool payloads with non-finite max_output_length values (NaN or infinities) previously escaped coercion as ValueError/OverflowError, and negative infinity could disable output truncation. This change bounds all non-finite numeric limits to zero while preserving integer limits, including arbitrarily large integers.

Test plan

• Added focused regressions for NaN, positive/negative infinity, and a 400-digit integer.
• tests/test_shell_call_serialization.py and tests/test_shell_tool.py: 49 passed.
• .agents/skills/code-change-verification/scripts/run.sh: format, lint, typecheck, and full tests passed.
• git diff --check: passed.

Issue number

Not linked to an issue; this is a directly reproducible malformed model-payload bug.

Checks

• I've added new tests, if relevant
• I've run .agents/skills/code-change-verification/scripts/run.sh
• I've confirmed all verification steps pass
• If using Codex, I've run /review before submitting this PR

[seratch]

On the current base, these values fail during normalization before the shell executor runs. This patch instead converts them to zero, executes the command, and silently discards its output, so it weakens the failure behavior rather than fixing a supported case.

I am going to close this PR. If a supported provider is observed emitting this payload, please provide a sanitized real response and its provider/version. The appropriate SDK behavior would then be an actionable ModelBehaviorError before execution, not coercion to a valid zero limit.