feat: read recipes from stdin and add --tasks-format

[josegonzalez]

apply, plan, and validate now accept - as the recipe source, spelled either as --tasks - or as a bare positional argument, matching the docket fmt - that already existed. The new --tasks-format flag states the recipe format outright instead of inferring it from the file extension, which matters for stdin, for an extension that is absent or misleading, and for a flow-style YAML recipe that would otherwise be sniffed as JSON5. tasks-format joins the reserved input names.

A piped recipe behaves like any other: its inputs: still become --<name> flags, and it takes precedence over a tasks.yml in the working directory. This makes docket export --output - | docket apply - work end to end.

The flag is named --tasks-format rather than --format for two reasons. --format already means output format elsewhere in the Dokku ecosystem, sitting next to docket's own --json, and reserving the un-hyphenated format would have broken any recipe declaring an input by that name. A hyphenated input name is already rejected, so reserving tasks-format cannot break an existing recipe. This also keeps plain --format free for init and export, which infer their output format from --output and today can only emit YAML to stdout.

Closes the first half of #406. Two follow-ups are carved out of the rest: #409 for the ansible-dokku wrapper developer documentation, and #410 for a format override on init and export output.