Remove leftover code and split long compiler functions

[itsfuad]

Summary

• What changed: Removed code that lost its last caller in earlier redesigns, removed parameters and results nobody uses, merged several duplicated blocks, and split four long functions into named steps.
• Why this change is needed: A whole-compiler clean-code audit found dead code, repeated logic and very long functions that made the code harder to read and change.
• Owning phase/package/component: Parser, lexer, binder, resolver, typechecker, place, flow/ownership analysis, THIR, MIR lowering, LLVM backend, diagnostics, LSP, CLI, pkg/toml, pkg/manifest, pkg/distribution, pkg/registry.
• Scope boundaries / non-goals: No language, diagnostic or output change. Not included: nil-receiver guard cleanup, parser table rewrite, moving ParseIntegerBuiltin or CompilerOp, and the output-ordering issue described under Risks.

Depends on #160; retarget to main after that PR merges.

Production ownership

• Current non-test production consumer(s) of each new production symbol:
  • (*thir.Match).Arm: analysis/ownership.go, analysis/flow_thir.go.
  • (*lowerer).storeInTemp: 19 cases in mir/expr_lower.go.
  • scanQuoted: three call sites in diagnostics/syntax_highlighter.go.
  • stringHandler(prefixLen, kind): two lexer patterns.
  • handleRequest: hover, definition, completion and rename in the LSP loop.
  • Step functions split out of GenerateLLVMIR, checkForInStmt, workspaceIndex.rebuild and Run: one caller each, the function they came from.
• Execution path showing where the new behavior is reached in production: No new behavior. The same paths run through the renamed or extracted functions.
• Existing implementation reused, extended, replaced, or intentionally left unchanged: THIR-based origin resolution in analysis/flow_thir.go remains the only resolver; the old AST-based one in place is removed. peeper remove keeps its own logic in cmd/cli/remove.go.
• Wrappers, aliases, duplicate paths, compatibility shims, or experimental scaffolding removed: private callableType (merged into CallableTypeFor), place.Binding, typechecker copies of two flow diagnostics, cstringHandler, bindModuleScopeType, bindModuleScopeTypeIfUnset, three copies of the quote scanner, four copies of the LSP request block, the parser context stack.
• Any new production symbol without a current production consumer: None new. Emitter.printUnderlineIndent, toml.Data.HasSection, toml.Data.EnsureSection and toml.ParseBytes already had no caller; they are kept at maintainer request and now carry a comment saying so.

Design and behavior

• Previous behavior: Unchanged.
• New behavior: Unchanged.
• Key invariants preserved: Operand evaluation order and temp numbering in MIR lowering; LLVM text and value numbering; diagnostic text, codes and labels; zip timestamps (still UTC 1980-01-01); lazy creation of project contexts in the LSP workspace scan.
• Intentional behavioral changes: None.
• Error / edge-case handling: reportUnresolved keeps its nil-node and nil-bag guard. tar.TypeRegA was dropped from one case; the Go tar reader already maps it to TypeReg.
• Why this implementation belongs in this layer rather than another phase/package: Each helper lives beside the code it replaced. Arm sits on thir.Match because its callers moved to THIR.

Validation

gofmt -l cmd internal pkg scripts
(no output)

go build ./... && go vet ./...
(no output)

go test ./...
52 packages ok, 0 failures

go run ./scripts/bundle.go
exit 0

PEEPER_BIN=$PWD/build/bin/peeper go test ./x_test/ -count=1
ok  compiler/x_test

go run mvdan.cc/unparam@latest ./...
(no findings)

go run honnef.co/go/tools/cmd/staticcheck@latest ./...
production code: only printUnderlineIndent (kept on purpose)

git diff --check
(no output)

• Targeted tests added or updated: cmd/cli tests call GetCommand(nil) instead of the removed installAllDependencies; place tests use the new resolver signature; ServerState.recompile moved to internal/lsp/state_test.go.
• Existing regression coverage exercised: Full Go suite and all x_test fixtures.
• Manual verification, if applicable: Built all 288 x_test projects with -keep-gen before and after. 249 .ll files (compared with line order ignored) and all stdout/stderr are identical. Ran peeper remove on a scratch project.
• Checks not run: Cross-target builds (Linux/386, Windows, macOS) were not run separately; no target-specific code changed.

Diff sanity

• Unrelated production changes: pkg/distribution/pack.go also turns two if / else if chains into switch. Same behavior.
• Duplicated logic introduced: None.
• Dead or unreachable code introduced: None.
• Temporary/debug code remaining: None.
• Comments/docs that became stale because of this change: Updated docs/architecture/cli-and-packages.md, frontend.md and semantics-bindings.md. Code-tour.md section 12 still shows the instruction switch inside GenerateLLVMIR; it now lives in emitBlock. The snippet is marked "simplified" and was left as is.

Risks and follow-up

• Known risks or assumptions: Eight tests in internal/semantics/place/origin_test.go were deleted with the AST-based resolver they covered. Nothing in production called that resolver since 2026-09-21.
• Behavior intentionally changed: None.
• Behavior intentionally not changed: analysis/effects.go keeps its own arm loop: it visits every arm with a case, not only the first, which can matter for programs with duplicate arms.
• Remaining gaps: Found during validation and not fixed here: two identical compiler runs can differ in MIR type numbering and in the order of LLVM declare lines. Content is the same when line order is ignored.
• Linked issues / follow-up work: None linked.
• Anything that should block merge: None.