Says which use options the attempt cap ignores

[johnnyt]

Summary

The README's "Capping a handler's attempts" section said any :max_attempts value other than a positive integer fails the handler's compile, and the StatifierOban.Invoke.Handler moduledoc sentence just before its literal-options paragraph said the same. Both now match that paragraph:

• README: the options must be a literal keyword list in the use; a keyword list reaching it through a module attribute, a variable or a function call is ignored whole, as is any argument that is not a keyword list; the one key read is :max_attempts and any other key is ignored; max_attempts: nil declares no cap; none of these is an error.
• Moduledoc: the sentence now says any other value fails the compile except in the cases the next paragraph lists, none of which is an error.

No other README section changes. Docs only; no behaviour changes, so no changelog fragment (changelog.d/README.md excludes documentation).

Review (in-turn, gate tier)

Each claim was checked against lib/statifier_oban/invoke/handler.ex by anchor: __using__/1 reads :max_attempts only when Keyword.keyword?(opts) holds on the use argument as written, so an attribute, a variable, a call or a non-list argument yields no cap, and Keyword.get/2 reads only that key; __max_attempts__!/2 answers nil for nil and raises ArgumentError for any value that is not a positive integer; put_max_attempts/2 adds nothing to the job when the handler's cap is nil. The misspelled-key and non-literal cases are pinned by the tests "an unknown key in the use options compiles clean and stores no cap" and "use options that are not a literal keyword list compile and store no cap" in test/statifier_oban/invoke/handler_test.exs; the bad-value raise by "a cap that is not a positive integer fails the handler's compile". No test compiles a handler with an explicit max_attempts: nil; that gap is noted for follow-up, not changed here.

Gate

The edit is in lib/, a build path, so the full mix quality ran green on the rebased head (313 of 313 tests).

Refs: sob-tfq