SymbolTableBuilder crashes on relative project_dir: eager relative_to() in _add_class next() default

[rahlk]

Pre-existing bug on main (verified byte-identical at 2cab542), discovered while verifying the #80 fix branch: constructing SymbolTableBuilder with a relative project_dir crashes analysis of any file containing a class.

Symptom

SymbolTableBuilder(Path("test/fixtures/single_functionalities/method_call_resolution"), None) \
 .build_pymodule_from_file(Path("test/fixtures/.../main.py"))
# ValueError: '/abs/path/.../main.py' is not in the subpath of 'test/fixtures/...'

Root cause

In _add_class (codeanalyzer/syntactic_analysis/symbol_table_builder.py:202-209 on main), the Jedi-goto fallback signature is passed as next()'s default argument, so it is computed eagerly even when Jedi succeeds:

signature = next(
 (d.full_name for d in definitions if d.type == "class"),
 f"{Path(script.path).relative_to(self.project_dir)...}.{class_name}"
)

script.path is always absolute (Jedi resolves it), so relative_to(self.project_dir) raises ValueError whenever project_dir is relative. The except Exception handler below doesn't save it: it recomputes the identical expression and raises uncaught.

Fix direction

Normalize once in __init__: self.project_dir = Path(project_dir).resolve(). (Alternatively make the fallback lazy, but resolving at the boundary fixes every downstream relative_to in one place, including _callables.) Regression test: build a module via a builder constructed with a relative path.

Every current in-repo caller happens to pass resolved paths (e.g. test/conftest.py fixtures), which is why the suite never hits it — library consumers won't be so lucky.