feat[next]: name program and static args in "no program compiled" error

[havogt]

Description

Draft — stacked on #2733, which is the actual fix. Split out of #2731 (now closed) because this
part is a message improvement and can be dropped independently.

When no compiled variant matches a call, the error said only

RuntimeError: No program compiled for this set of static arguments.

— neither which program, nor which arguments, nor that the offset provider takes part in the
selection. In a model with hundreds of programs that is not enough to act on.

Changes

RuntimeError: No program compiled for this set of static arguments of 'prog': cond=False. Note
that a variant is also selected by the identity of the 'offset_provider' entries and, for
generic programs, by the argument types.

_describe_argument_descriptors renders the key's descriptor values against the expressions
they were extracted from, so it covers both StaticArg (cond=False) and
FieldDomainDescriptor ((out).domain=Domain(...)), and yields an empty string when the pool
has no argument descriptors at all.

It deliberately does not use zip(..., strict=True): this runs while building an error
message, so a length mismatch must degrade to a shorter description rather than replace the
error the user actually needs — which is the failure mode #2733 exists to remove.

Open question

Is naming the arguments worth the ~15 lines? The "Note that ..." sentence about
offset_provider identity carries most of the debugging value on its own and would be a
two-line change.

Requirements

• All fixes and/or new features come with corresponding tests.
  Three unit tests in tests/next_tests/unit_tests/otf_tests/test_compiled_program.py:
  static arg, static domain, and no argument descriptors.
• Important design decisions have been documented in the appropriate ADR.