support pass extensions for plan logging and profiling per optimization pass

[yliang412]

Problem

The optd rewrite and optimization phases were hard-coded in Cascades::optimize, which made them difficult to observe or instrument. EXPLAIN only showed a limited before/after view, and there was no systematic way to collect per-phase plan snapshots or timing metrics.

Summary of changes

• Added a generic PlanPass / PassExtension / PassManager framework in optd-core.
• Converted decorrelation and simplification into named PlanPass implementations.
• Added async phase support in PassManager so the memo-based cascades phase can use the same extension hooks.
• Wired Cascades through PassManager, including reverse-order after_pass callbacks so extensions behave like proper around hooks.
• Added ExplainPassExtension in the DataFusion planner to record optd plans after each named phase.
• Added physical_plan after optd-cascades to EXPLAIN output and removed the redundant physical_plan after optd-finalized row.
• Updated planner test snapshot filtering to include all logical_plan after optd-* and physical_plan after optd-* sections.
• Added runtime pass profiling via SET optd.profile_passes = true.
• Moved profiling into optd/core/src/rules/profile.rs so tracing uses the default optd_core::rules::profile target.
• Emitted profiling metrics through tracing with structured fields for pass name, changed flag, elapsed time, total time, and invocation count.
• Regenerated planner snapshots to include the new per-phase explain sections.

Example: Profiling Tracing Output

RUST_LOG=optd_core::rules::profile=info cargo run --bin planner_test_apply -- tpch::q3 -- --exact
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 0.18s
     Running `target/debug/planner_test_apply 'tpch::q3' -- --exact`
2026-04-28T16:55:39.305964Z  INFO optd_core::rules::profile: optimizer pass profile pass="decorrelation" changed=false elapsed_ms=0.0179 total_ms=0.0179 invocations=1
2026-04-28T16:55:39.306993Z  INFO optd_core::rules::profile: optimizer pass profile pass="simplification" changed=true elapsed_ms=0.473 total_ms=0.473 invocations=1
2026-04-28T16:55:39.309540Z  INFO optd_core::rules::profile: optimizer pass profile pass="cascades" changed=true elapsed_ms=2.25 total_ms=2.25 invocations=1

(The log actually get print twice, because we plan the query twice, once for planning, once for execution)

Future work

• Consider whether cascades should become a first-class pass abstraction instead of using async phase hooks.