Match float literals to float32 comparisons

[deepakganesh78]

Fixes #609

Root cause

Expr compiles floating-point literals as float64. Comparing one directly with a float32 operand therefore compares the rounded float32 value with the original float64 literal, unlike Go's conversion of the untyped literal to float32.

Fix

After type checking, patch float literals used directly in comparisons with float32 operands into typed float32 constants. This handles either operand order and signed literals without changing mixed float32/float64 arithmetic.

Tests

• go test .\checker . -run 'Test(Float32LiteralComparisons|Check)' -count=1
• go test ./...
• go test -tags=expr_debug -run=TestDebugger -v ./vm
• go vet ./...

The regression test covers optimized and unoptimized compilation, both operand orders, signed literals, relational and inequality comparisons, and verifies float64 behavior remains unchanged.

go test -race . was not available on this Windows host because the installed Go toolchain has CGO disabled.

[sanmaxdev]

Verified this fixes #609. Reproduced F32 == 12.34 returning false on master, and it returns true on this branch. The patcher approach is what was requested in the issue thread, and it correctly handles both operand orders plus signed literals without touching mixed float32/float64 arithmetic.

Verification run on the PR head:

• go test ./... (full suite, no failures)
• go vet ./... (clean)
• go test -race -count=1 ./checker . (green)

One test gap: isComparisonOperator returns true for <, >, and >=, but the table in TestFloat32LiteralComparisons only exercises ==, !=, and <=. Adding rows for the remaining three would cover every branch of the helper and guard against a future regression that narrows the accepted operator set.