feat(serverless): address an environment on invoke and get_task (RUNSERV-1238) - #5

RUNSERV-1238: an app id now names one copy per environment, so the client has to say which one it means.

• Runware(serverless_environment="staging") sets it for the client; invoke(..., environment=) and get_task(..., environment=) override it per call.
• Unset sends no environment parameter, so the platform reads production and existing callers send exactly the same requests as before.
• Waiting polls in the invocation's environment. The task body carries no environment, so a poll without one would read production, 404, and fail once the 30s not-found grace runs out. The poll-timeout hint names the environment too.
• Names are checked against the API's pattern before any request; a bad serverless_environment fails when the client is built.
• An environment paused at its monthly budget (RUNSERV-1240) answers invoke with a 429 that holds until the budget is raised or the month turns. A 429 whose problem type is https://docs.runware.ai/serverless/errors#environment-paused is now a quota error raised at once, not retried and not slept on its Retry-After; any other 429 is retried as before. The pause PR must use that type.

The SDK has no deploy or app reads yet, so invoke and get_task are the whole surface this touches.

Release note: the prod API (as of bf039685) predates environments and ignores the parameter, which means a non-production call would reach the production copy. Publish only once prod runs Runware/serverless#787 or later; dev already does. Merge after the RUNSERV-1240 pause work, so the paused type above is live.

Tested: ruff, pyright, pytest (360 passed), and a local run against the serverless API from main with production and staging copies of one app: each copy's tasks ran on its own consumer and were readable only in their own environment. Against the prod API commit, environment=staging ran on the production copy, hence the release note.