fix: support model prefix in with_dimensions/with_measures after joins

[boringdata]

Summary

• Support model prefix navigation (t.flights.carrier) in with_dimensions() and with_measures() after joins (Inconsistent model prefix behavior in lambdas: works in filter() but fails in with_dimensions/with_measures #136)
• Previously only .filter() supported this pattern
• Adds _ColumnPrefixProxy to ColumnScope/MeasureScope and _DimensionTableProxy for dimension evaluation

Test plan

• test_model_prefix_in_with_dimensions — t.carriers.name works in with_dimensions after join
• All existing join prefixing tests pass
• All existing deferred API tests pass

🤖 Generated with Claude Code

[boringdata]

Codex Review Summary

Codex (gpt-5.3-codex) reviewed this PR and found 2 P1 issues related to silent proxy fallbacks. Both have been fixed in commit bb07a6f:

Issues Found & Fixed

• [P1] _ColumnPrefixProxy.__getattr__ (measure_scope.py:34-38): Was falling back to getattr(self._tbl, name) when prefixed column not found, which could silently resolve t.customers.id to unprefixed t.id. Fixed: Now raises AttributeError with available prefixed columns listed.

• [P1] _DimPrefixProxy.__getattr__ (ops.py:862-866): Same silent fallback issue for dimension lookups. Fixed: Now raises AttributeError listing available prefixed dimensions.

• [P2] Dimension.__call__ proxy error swallowing (ops.py:918-919): Retry path was catching all exceptions from proxy evaluation, masking dependency resolution errors. Fixed: Now re-raises AttributeError from proxy for proper dependency chain detection.

All 103 tests passing after fixes.