Remove eps_attr config field

[RecreationalMath]

Description

Closes #1376.

Removes the unused eps_attr field from TransformerBridgeConfig. It was set by 8 architecture adapters but read nowhere (its only consumer was deleted in #1119 when NormalizationBridge moved to direct HF delegation), so removing it changes no model behavior.

It is a breaking API change: the eps_attr constructor argument and the config.eps_attr attribute are gone (the field was live, never deprecated). Serialized configs are unaffected (from_dict drops unknown keys). The break only hits code that passes eps_attr= to the constructor or reads config.eps_attr.

Removed in one sweep: the config field, the 8 adapter self.cfg.eps_attr = ... assignments, the 6 adapter-test assertions, and the doc/template mentions (including the now-dead cfg.eps_attr troubleshooting row in the numerical-divergence guide).

Note: deleted the troubleshooting row rather than rewire it, since the eps machinery already auto-introspects .eps/.variance_epsilon (a transformers 5.8.1 scan found no HF norm using another attribute name), so the symptom is not user-actionable via this field.

Type of change

• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• This change requires a documentation update

Checklist:

• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• New and existing unit tests pass locally with my changes
• I have not rewritten tests relating to key interfaces which would affect backward compatibility