Fix Firstrade buying-power sizing and dry-run isolation

[Pigbibi]

Why

• Firstrade rejected an order estimate because the requested amount consumed the reported buying power with no operational reserve for unsettled funds and price movement.
• Dry-run execution could still mutate durable strategy state.
• Runtime reconciliation and monitor handling needed to distinguish safe validation from live execution.

What changed

• Apply the configured reserved-cash ratio when sizing executable orders and expose it through the deployment sync contract.
• Prevent dry-run routes from persisting live strategy state.
• Harden runtime target reconciliation, monitor preservation, and broker adapter error handling.
• Add regression coverage for buying-power reservation, dry-run isolation, Cloud Run routes, and workflow reconciliation.

Verification

• pytest: 202 passed.
• ruff check: passed for all changed Python files.
• actionlint: passed for the changed workflow.
• git diff --check: passed.
• Codex review (not AIAudit): no actionable findings.

Production safety

• A 1% cash reserve is already configured on the live service as a mitigation.
• No Firstrade dry-run or live order endpoint was invoked after applying the mitigation.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7abf81daad

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid retaining the dispatcher beside direct monitor jobs

When SYNC_PLAN_JSON contains multiple targets and DIRECT_MONITOR_MIGRATION_COMPLETE=true, the scheduler step still creates and resumes the selected service's direct probe and precheck jobs, but this single-target condition forces direct_jobs_exist to remain false and preserves the global dispatcher. Because the preceding environment sync configures MONITOR_DISPATCH_TARGETS_JSON with that same selected service, the dispatcher continues invoking its due monitors alongside the direct jobs, causing duplicate dry-run previews, session checks, and failure notifications. Either remove migrated targets from the dispatcher configuration or retire the dispatcher after all represented targets have direct jobs.

Useful? React with 👍 / 👎.