op-reth: assert builder_config shares its live interop failsafe handle

[op-will]

What

Adds liveness tests for two OpBuilderConfig handles that OpNode::builder_config() threads through: the interop failsafe and the SDM operator opt-in.

Why

builder_config_reflects_node_fields sets each node field before calling the accessor, so it proves values propagate — but not that the Arc-backed handles stay shared. Of the three shared handles, only da_config had a liveness test (builder_config_da_is_live_shared_handle).

That gap matters most for the interop failsafe, because of when it is written. The filter client sets it at runtime, long after the payload builder has been constructed. So a refactor that copied the flag's value instead of sharing the handle — say interop_failsafe: InteropFailsafe::default(), or a plain bool — would leave the builder permanently ungated, and every existing assertion would still pass. Interop transactions would be included during a failsafe.

The same reasoning applies to operator_sdm_opt_in, which the admin RPC flips after startup.

How

Mirrors the existing DA case exactly: take the config first, flip the node's handle after, assert the config observes the change. The failsafe test additionally clears the flag, proving a single shared handle rather than a one-way latch.

Verification

• All 4 builder_config tests pass.
• Mutation-tested: replacing the accessor's interop_failsafe with InteropFailsafe::default() makes the new test fail with its intended message (builder_config must carry the node's live interop failsafe handle, not a detached copy), confirming it detects the regression rather than passing vacuously.
• cargo +nightly fmt clean.

Tests only — no production code touched.

🤖 Generated with Claude Code