feat(dsl): scena-check, the DSL checking CLI (p7-s5)

[JArmandoAnaya]

First half of slice 43i. tools/scena-check — load a .osc file, follow its
imports, report what the checker finds.

$ scena-check my_scenario.osc
my_scenario.osc: ok, 412 types across 3 files

A thin consumer of scena::dsl::check_file with no language semantics of its
own. It links the DSL frontend alone: checking a DSL file needs no engine, no
road backend and no XML frontend.

Exit codes

0 ok, 2 usage, 3 the source did not check, 4 the input could not be
read.

3 and 4 are deliberately distinct. Scena's status model separates a defect
in the content from the host handing the library something unusable, and the
exit codes follow that line — a file full of type errors is the scenario's
problem, an unreadable path is the invocation's, and a build script can branch
on which.

A diagnostic gap the CLI made obvious

Resolver and expression diagnostics carried a line and a column but no file.
That is fine inside a test that checks one string, and useless in a checker: a
Program spans every file its root imported, so a diagnostic from an imported
file was indistinguishable from one in the file you passed.

Scope and ExpressionContext now carry the source file; the three resolver
emitters and expression typing's fail stamp it. dsl_import_test.cpp gains
ADiagnosticNamesTheFileItCameFrom, which builds a two-file program with an
error in each and asserts both that no diagnostic has an empty file and that
each names the right one.

This is why the slice was worth splitting — the gap is only visible once
something has to print the diagnostics for a human.

Tests

18 subprocess cases, reusing scena-run's harness with its portability fixes
intact: the cmd.exe outer-quote wrapping and the POSIX exit-status unpacking.
They cover every exit code, the diagnostic shape and position, search-path
ordering (two copies of a module, resolved differently depending on -I
order), file and module imports, --strict promoting a warning, --quiet
staying silent while still exiting non-zero, --no-standard-library, the three
usage errors, multi-diagnostic recovery, and byte-identical output across two
runs.

One pins something specific to this frontend: a DSL diagnostic prints no
bracketed rule id, because the DSL standard defines no asam.net: rule
identifiers. The section reference in the message is the citation.

1304 gtest (+19: 18 CLI + 1 import), 184 pytest.

Docs: docs/user-guide/scena-check.md and an index entry.

What is left on #43

43i-2: the C ABI scn_check_dsl, the Python check_dsl, the parity audit
update and an example. That PR carries Closes #43.

Refs #43