fix(serverless): default workersStandby to workersMin so workers=(0, N) scales to zero

[justinwlin]

Summary

workers=(min, max) did not set workersStandby when building the saveEndpoint create/update payload — the field was simply absent from the model, so the RunPod API applied its own server-side default. Deploying with workers=(0, 1) produced workersMin: 0 and workersStandby: 1: the endpoint kept one warm worker forever and never scaled to zero, invisibly. This PR defaults workersStandby to workersMin (with explicit-override support), so the deployed config matches the requested scaling range.

Fixes #364

Internal: CON-1229

What changed

• src/runpod_flash/core/resources/serverless.py — new optional workersStandby field on ServerlessResource; a model validator defaults it to workersMin when unset and validates explicit values against the (workersMin, workersMax) range. Because the deploy/update payloads are built from model_dump(), both create and update now send workersStandby.
• src/runpod_flash/endpoint.py — new keyword-only workers_standby param on Endpoint, range-validated and passed through to the resource config (issue suggestion ux update to terminal #2). When omitted, the resource default (workersMin) applies.
• docs/Flash_SDK_Reference.md — document workers_standby and the standby-follows-min behavior next to workers.
• tests/unit/resources/test_serverless.py, tests/unit/test_endpoint.py — 12 new tests: workers=(0, 1) yields workersMin: 0 / workersStandby: 0 in the deploy payload, workers=(2, 5) yields workersStandby: 2, explicit overrides respected, out-of-range values raise.

Notes

• workersStandby is intentionally not in _hashed_fields, so resource_id identity is unchanged. It does enter the exclude-based config_hash; for legacy pickled resources the field is simply absent (verified config_hash/model_dump don't error), so the first deploy after upgrade detects drift once and issues a non-structural saveEndpoint update that corrects standby on existing endpoints.
• LB resources keep their existing behavior via workersMin=1 defaults.

How verified

• pytest tests/unit/test_endpoint.py tests/unit/resources/test_serverless.py tests/unit/resources/test_live_serverless.py — 342 passed.
• Full pytest tests/unit — 2570 passed, 3 failed; the same 3 tests fail identically at the base commit (unrelated pre-existing test-ordering issues in test_load_balancer_sls_stub.py / test_regressions.py; they pass in isolation).
• ruff format + ruff check — clean. mypy . — same 260 pre-existing errors as the base commit (error set identical modulo line shifts).
• Simulated a legacy unpickled ServerlessResource (no workersStandby in instance state): config_hash, resource_id, and model_dump all work.

Confirmed via mocked payloads only — no live RunPod deploy was run.