fix(subagent): honor configured and requested step budgets

[yanmxa]

What & why

Agent(max_steps=1) currently runs with at least 500 inference steps. The executor treats 500 as a minimum and only lets the request increase the definition's budget, so short delegated tasks cannot enforce their requested limit.

Honor positive agent-definition budgets and let a positive request override them, including smaller values. Use 500 only when no positive budget is provided. Update the tool schema and authoring guide to match. This is independent of the reporting and display fixes in #614.

Type of change

• Bug fix
• Docs

Testing

GOCACHE=/private/tmp/san-go-build-cache GOLANGCI_LINT_CACHE=/private/tmp/san-budget-lint-cache.42E9pI make ci

The checks include formatting, full build, vet, lint, layer checks, and the full unit/integration suite with the race detector. Configuration regressions cover smaller request and definition budgets. A real runtime regression proves MaxSteps=1 performs exactly one inference.

Checklist

• Follows Conventional Commits
• Every commit signed off
• Docs updated
• Rebased onto latest upstream/main

[san-ci]

[REVIEW NOTIFIER] This PR requires a review from @zhujian7.

Once this PR has been reviewed and has the lgtm label, it will be automatically squash-merged.

Suggested reviewer

Based on OWNERS, @zhujian7 is suggested to review this PR.

How to get your PR reviewed:

• Address reviewer feedback and push changes
• Reviewer adds lgtm label when satisfied
• An approving review + lgtm will trigger auto-merge

[zhujian7]

/lgtm

[san-ci]

⛔ Auto-merge failed

San-CI attempted to auto-merge this PR but the merge was rejected by GitHub.

Reason: 2 of 2 required status checks are in progress. - https://docs.github.com/articles/about-protected-branches

Common causes:

• Branch protection rules are not fully satisfied
• Required conversations are not resolved
• The PR is out of date with the base branch
• Required reviews or status checks are missing

Once the issue is resolved, San-CI will retry on the next relevant event.