Remove constructors from merge functions

[saulshanabrook]

This removes constructor-producing encodings from merge functions so merge results are eager values selected from the existing inputs, which is needed for proof mode and for avoiding merge-driven e-node growth.

Stack

PR against Oliver's oflatt-latest-egglog branch.

This now depends on:

• egglog combined branch: egraphs-good/egglog, branch codex/pr927-plus-pr930-eggcc-proof (9b5ce111), which merges container PR #927 with PR #930 and the proof-mode command-macro type-info fix.
• egglog-experimental helper PR: [codex] Add eggcc proof baseline helpers egglog-experimental#56 (19ed866).

The Cargo manifest points at the experimental PR branch, and Cargo.lock pins that branch plus the combined egglog branch.

What changed

• TermAndCost is now a Pair Term i64.
• ExtractedExpr uses pair-min-by-second-i64, which returns one of the existing pair values directly.
• interval Bound is now Maybe (Either i64 bool).
• Lower/upper bound merges use maybe-either-i64-bool-max and maybe-either-i64-bool-min.
• Eggcc-specific aliases like IntB, BoolB, Dead, bound-int, and bound-bool were removed; eggcc now spells the generic container representation directly.
• IVTRes is now Pair IVTPayload i64, with merges using pair-min-by-second-i64.
• Rules that consume pair-backed values use pair-first and pair-second instead of matching on pair-shaped LHS terms.
• --run-mode egglog emits the generated egglog program for direct use with the experimental CLI.

Validation

Passed locally:

cargo check --locked
cargo test -p dag_in_context interval_analysis
cargo test -p dag_in_context optimizations::ivt
cargo test -p dag_in_context optimizations::memory
cargo test -p dag_in_context optimizations::switch_rewrites
cargo run --release -- --run-mode egglog --stop-after-n-passes 1 benchmarks/passing/polybench/linear-algebra/kernels/2mm.bril > .tmp/proof-profile/2mm-pass1-current.egg
/Users/saul/p/wt/egglog-experimental/eggcc-proof-baseline-pr927/target/release/egglog-experimental .tmp/proof-profile/2mm-pass1-current.egg

Current composed-stack validation:

# eggcc, after lockfile repin
cargo check
cargo check --locked

# egglog-experimental PR #56, release CLI against full checked-in 2mm export
target/release/egglog-experimental --proofs tests/fixtures/eggcc-2mm-pass1-merge-old.egg

The full 2mm pass-1 export is checked into egglog-experimental PR #56 as tests/fixtures/eggcc-2mm-pass1-merge-old.egg; it contains no :no-merge and runs successfully in proof mode with the release CLI. The default experimental test suite keeps the full-export canary ignored because debug-profile proof execution was too slow for CI.

Performance

Earlier local 2mm export runs for the constructor encoding vs eager encoding showed the eager encoding faster (about 13% on that run). I did not rerun the 5x benchmark after replacing eggcc-specific aliases with direct generic container forms.

[oflatt]

small comments from our discussion last time