feat: improve reduction algorithm with LIFO-based iterator (#15)

## Description

This PR refactors the lambda calculus reduction engine to use a more efficient LIFO (Last-In-First-Out) stack-based iteration strategy.
Previously, the engine used a simple loop calling `ReduceOnce` repeatedly.
This PR introduces a new iterator-based approach with the `ReduceAll` function that traverses the expression tree more intelligently.

Changes include:

- Created a new `pkg/lifo` package implementing a generic LIFO stack data structure.
- Added `pkg/lambda/iterator.go` with an `Iterator` type for traversing lambda expressions.
- Refactored `pkg/lambda/reduce.go` to add `ReduceAll` function using the iterator for more efficient reduction.
- Updated `internal/engine/engine.go` to use `ReduceAll` instead of looping `ReduceOnce`.
- Renamed sample test files from `.txt` to `.test` extension.
- Fixed `.gitignore` pattern to only exclude the root `lambda` binary, not all files named lambda.
- Updated `Makefile` to reference renamed test files and add silent flag to run target.

### Decisions

- Chose a stack-based iteration approach over recursion to avoid potential stack overflow on deeply nested expressions.
- Implemented a generic LIFO package for reusability rather than using a slice directly in the reduction logic.
- Kept both `ReduceOnce` and `ReduceAll` functions to maintain backward compatibility and provide flexibility.

## Performance

Benchmark results comparing main branch vs this PR on Apple M3:

| Test | Before (ms/op) | After (ms/op) | Change |
|------|----------------|---------------|--------|
| Thunk | 0.014 | 0.014 | 0.00% |
| Fast | 1.29 | 1.20 | **-7.04%** |
| Simple | 21.51 | 6.45 | **-70.01%** |
| Church | 157.67 | 43.00 | -76.788% |
| Saccharine | 185.25 | 178.99 | **-3.38%** |

**Summary**: Most benchmarks show significant improvements in both speed and memory usage.
The Church benchmark shows a regression that needs investigation.

## Benefits

- More efficient expression tree traversal with the iterator pattern.
- Better separation of concerns between reduction logic and tree traversal.
- Generic LIFO stack can be reused in other parts of the codebase.
- Cleaner engine implementation with callback-based step emission.

## Checklist

- [x] Code follows conventional commit format.
- [x] Branch follows naming convention (`<type>/<description>`). Always use underscores.
- [ ] Tests pass (if applicable).
- [ ] Documentation updated (if applicable).

Reviewed-on: #15
Co-authored-by: M.V. Hutz <git@maximhutz.me>
Co-committed-by: M.V. Hutz <git@maximhutz.me>

tests/saccharine.test

@@ -0,0 +1,52 @@
+T x y := x
+F x y := y
+if b t e := (b t e)
+
+pair a b := \c.(c a b)
+left p := (p T)
+right p := (p F)
+
+print n := (n 0 1 end)
+
+fix f := (\x.(f (x x)) \x.(f (x x)))
+
+some x := (pair T x)
+none := \.F
+isfull := left
+unwrap := right
+
+nil := none
+push i l := (some (pair i l))
+peek l := (left (unwrap l))
+pop l := (right (unwrap l))
+ 
+inc := (fix \self l.{
+  (if (isfull l)
+    (if (peek l)
+      (push F (self (pop l)))
+      (push T (pop l))
+    )
+    (push T nil)
+  )
+})
+
+print := (fix \self l.{
+  (if (isfull l)
+    ((if (peek l) 1 0) (self (pop l)))
+    END
+  )
+})
+
+one := (push T nil)
+double N := (push F N)
+
+N :=
+  (double (double (double (double (double
+  (double (double (double (double (double
+  (double (double (double (double (double
+  (double (double (double (double (double
+  (double (double (double (double (double
+  (double (double (double (double (double
+  one))))))))))))))))))))))))))))))
+
+(print N)