fix: late constant folding for cross-module if-statement DCE

[Felipeness]

Summary

When bundling with --minify-syntax, imported constants like export const DEV = false are inlined at print time via ConstValues. However, if statements (SIf) with these inlined constants were not folded — only if expressions (EIf/ternary) were. This left dead branches in the output:

// constants.js
export const DEV = false

// entry.js
import { DEV } from './constants'
if (DEV) {
  let detail = "detailed: " + msg
  throw new Error(detail)
} else {
  throw new Error("ERROR_CODE")
}

Before:

// Both branches kept, condition becomes !1 but if-statement stays
if (!1) { let e = "detailed: " + msg; throw new Error(e); } else throw new Error("ERROR_CODE");

After:

// Dead branch eliminated, only live branch remains
throw new Error("ERROR_CODE");

Approach

Added late constant folding for SIf in printStmt() (mirroring how EIf is already handled in lateConstantFoldUnaryOrBinaryOrIfExpr()):

1. When MinifySyntax is enabled, fold the test expression through lateConstantFoldUnaryOrBinaryOrIfExpr() (which resolves EImportIdentifier refs via ConstValues)
2. Use ToBooleanWithSideEffects() to determine if the condition is statically known
3. If true → print only the "yes" branch; if false → print only the "no" branch (or nothing)

Also added isLateConstantFoldedSideEffectFree() to drop residual expression statements like !1; that result from logical expressions with inlined constants (e.g., DEV && console.log("debug") → !1;).

Test

go test ./... -count=1
ok  github.com/evanw/esbuild/internal/bundler_tests  2.210s
ok  github.com/evanw/esbuild/internal/js_printer      0.733s
... (all packages pass)

Three new snapshot tests added:

• TestCrossModuleConstantFoldingIfStmt — basic if (DEV) elimination
• TestCrossModuleConstantFoldingIfStmtThrow — if/else with throw in both branches
• TestCrossModuleConstantFoldingIfStmtMultiple — multiple independent if statements

Fixes #2589, fixes #3964.

[Felipeness]

Hey @evanw, just a gentle ping on this and the related #4412. Both PRs have passing tests (\FAIL ./... [setup failed]
FAIL) and address longstanding issues (#2589, #3964, #2605). Happy to adjust anything if the approach doesn't align with your vision for esbuild. No rush — I know you review in batches!

[care44cubs-max]

Yes

[care44cubs-max]

Yes