Stabilize wavediff output for buffered file2-only changes

[jeet-dekivadia]

Fixes #16.

Summary

This makes wavediff output deterministic for the file2-only rows that were buffered while reading ahead in file2.

The comparison was already detecting the mismatch correctly, but that reporting path iterated buffered2, a HashMap<(time, handle), OwnedSignalValue>. That meant the same logical diff could render in hash iteration order. The patch collects those buffered rows, renders their names, sorts by (time, name, value), and then writes them out.

I also added a tiny VCD regression where file2 has three intermediate changes at time 10 and file1 has a later time-20 change. That forces the read-ahead buffer and checks the user-facing order:

10 t.a 1 (only in file2)
10 t.b 1 (only in file2)
10 t.c 1 (only in file2)

Validation

cargo build --release
cargo test
cargo clippy -- -D warnings
git diff --check
cargo run --quiet --bin wavediff -- tests/data/buffered_file2_base.vcd tests/data/buffered_file2_extra.vcd

The manual CLI check exits 1, as expected for a real waveform difference, and prints the three buffered file2-only rows in stable order.

[toddstrader]

I'd propose changing to something like this in order to avoid the new string cloning:
a68ad36

[jeet-dekivadia]

Thanks, agreed. I pushed 211e462 using that shape so the buffered values stay owned and the sorted rows carry an index instead of cloning the rendered value string per name.

Validation:

git diff --check
cargo test test_buffered_file2_only_diffs_are_sorted -- --nocapture
cargo run --quiet --bin wavediff -- tests/data/buffered_file2_base.vcd tests/data/buffered_file2_extra.vcd
cargo build --release
cargo test
cargo clippy -- -D warnings