Replace hardcoded list in get_all_scorers() with automatic discovery
using Python's __subclasses__() introspection. This eliminates the need
to manually maintain the scorer list when adding new scorers.

Key changes:
- Add _get_all_concrete_builtin_scorers() helper function that
  recursively discovers all concrete BuiltInScorer subclasses
- Update get_all_scorers() to use automatic discovery instead of
  hardcoded list
- Handle scorers requiring constructor args (e.g., Guidelines) by
  catching both TypeError and pydantic.ValidationError
- Remove unused is_databricks_uri import
- Update tests to expect 16+ scorers instead of 9/11
- Add test_builtin_scorer_discovery() to verify discovery mechanism

Benefits:
- Zero maintenance: new scorers automatically discovered
- Complete coverage: discovers all 16 instantiable scorers (vs 9/11)
- Adds 5+ scorers previously missing from hardcoded list: Fluency,
  Summarization, ConversationalSafety, ConversationalToolCallEfficiency,
  ConversationalRoleAdherence
- Future-proof: no code changes needed for new scorers

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Signed-off-by: Alkis Polyzotis <alkis.polyzotis@databricks.com>

Changes based on ALKIS comments:
- Move inspect import to top-level (no local imports)
- Consolidate tests: remove redundant test_get_all_scorers_oss
- Rewrite test to use public API get_all_scorers() instead of private
  _get_all_concrete_builtin_scorers()
- Improve test to verify exact set of expected scorers
- Simplify docstring and comments

The new test_get_all_scorers() is more comprehensive:
- Tests the public API directly (better encapsulation)
- Verifies exact count and exact set of scorers
- Explicitly checks Guidelines is excluded
- No dependency on tracking URI (simpler)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Signed-off-by: Alkis Polyzotis <alkis.polyzotis@databricks.com>

Address ALKIS feedback:
- Remove hardcoded count (expected_count = 16) in favor of directly
  comparing scorer_class_names with expected_scorers
- Remove all explanatory comments from test body
- Remove custom assertion messages - rely on pytest introspection
- Convert scorer_names check to simpler set comparison for duplicates

The test is now more concise and focuses purely on comparing the
actual set of discovered scorers against the expected set.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Signed-off-by: Alkis Polyzotis <alkis.polyzotis@databricks.com>

The test_list_builtin_scorers_shows_all_available_scorers test was
trying to patch is_databricks_uri which no longer exists in
builtin_scorers.py after our refactoring.

Since get_all_scorers() now returns all scorers regardless of
environment (no Databricks-specific conditional logic), the
parameterized test checking different behaviors is obsolete.

Simplified the test to verify that the CLI returns the same scorers
as get_all_scorers() without mocking or parameterization.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
Signed-off-by: Alkis Polyzotis <alkis.polyzotis@databricks.com>