Fix Terminator Issue

[willtebbutt]

Per discussions in #389 and #397 , this changes the normalisation pass to ensure that the CFG doesn't change. Another approach would have been to continue with how things are currently done and to properly reflect the changes in the CFG, but that would have been more work, and not obviously improved things further.

[codecov]

Codecov Report

Attention: Patch coverage is 90.90909% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/interpreter/ir_normalisation.jl	90.69%	4 Missing ⚠️

Files with missing lines	Coverage Δ
src/interpreter/ir_utils.jl	87.50% <100.00%> (ø)
src/interpreter/ir_normalisation.jl	90.21% <90.69%> (+0.14%)	⬆️

... and 1 file with indirect coverage changes

[github-actions]

Performance Ratio:
Ratio of time to compute gradient and time to compute function.
Warning: results are very approximate! See here for more context.

┌────────────────────────────┬──────────┬──────────┬─────────┬─────────────┬─────────┐
│                      Label │   Primal │ Mooncake │  Zygote │ ReverseDiff │  Enzyme │
│                     String │   String │   String │  String │      String │  String │
├────────────────────────────┼──────────┼──────────┼─────────┼─────────────┼─────────┤
│                   sum_1000 │ 100.0 ns │      1.8 │     1.1 │         5.5 │    8.21 │
│                  _sum_1000 │ 941.0 ns │     6.81 │  1570.0 │        36.6 │    1.07 │
│               sum_sin_1000 │  6.56 μs │     2.24 │    1.68 │        11.0 │    2.02 │
│              _sum_sin_1000 │   5.3 μs │     2.61 │   308.0 │        13.8 │    2.41 │
│                   kron_sum │ 343.0 μs │     39.8 │    3.69 │       201.0 │    8.72 │
│              kron_view_sum │ 338.0 μs │     43.0 │    9.71 │       217.0 │    98.9 │
│      naive_map_sin_cos_exp │  2.13 μs │      2.2 │ missing │        8.24 │    2.32 │
│            map_sin_cos_exp │  2.16 μs │     2.43 │    1.54 │        6.17 │    2.87 │
│      broadcast_sin_cos_exp │  2.21 μs │     2.31 │    2.31 │        1.48 │    2.25 │
│                 simple_mlp │ 427.0 μs │     4.61 │    2.96 │        7.28 │    3.37 │
│                     gp_lml │ 538.0 μs │     4.91 │    2.39 │     missing │    3.87 │
│ turing_broadcast_benchmark │  2.03 ms │     3.35 │ missing │        31.6 │ missing │
│         large_single_block │ 390.0 ns │     4.03 │  4420.0 │        30.7 │    2.18 │
└────────────────────────────┴──────────┴──────────┴─────────┴─────────────┴─────────┘

[willtebbutt]

It looks like this might have caused a small performance regression. I'm going to try a couple of things locally to resolve.

[willtebbutt]

Coverage failure is due to a chunk of code which is should only throw if something goes badly wrong -- everything else appears to be covered. I'm going to try to add a unit test for kill_edge!, but then I'll merge + tag.

[willtebbutt]

Unit test added. Will merge when CI (other than coverage) passes.