[Codegen] Move remaining pipelines to `iree_codegen` attrs (#24398)

Follow-up to the LLVMGPU, SPIRV, and LLVMCPU pipeline migrations
(#23816, #23851, #23864). Same goal: get backend-specific pipeline
identifiers out of the global `DispatchLoweringPassPipeline` enum and
behind `PipelineAttrInterface`.

This moves the remaining non-CPU/GPU pipeline cases to `iree_codegen`:
`#iree_codegen.vmvx_pipeline`,
`#iree_codegen.transform_dialect_codegen`, and
`#iree_codegen.no_pipeline`. It keeps custom pass pipelines represented
as `#iree_codegen.pass_pipeline<...>`.

These attrs intentionally stay in `iree_codegen`. `vmvx_pipeline` does
not move to a backend dialect because IREE does not have a VMVX codegen
dialect today; it can become a backend-owned enum attr if such a dialect
is introduced later. `transform_dialect_codegen` and `no_pipeline` are
also codegen-level sentinels, so modeling the remaining cases as unit
attrs avoids creating dialect surface just for pipeline spelling.

After this, `TranslationInfoAttr` and `iree_codegen.smt.constraints`
parse pipeline attributes directly through the interface. The old enum,
keyword compatibility parser, helper predicates, and textual parsing in
Python dialect tests can be removed.

Issue: https://github.com/iree-org/iree/issues/23535

Assisted-by: codex