self-host P3 slice 5: L3 and L4 as t27c runs them, and as the constitution states them under strict_laws (E_LAW) #8103

Part of #5980 (phase P3: "the semantic passes t27c runs before codegen (name resolution, type inference, L3/L4 checks)"). This is the fifth and last P3 slice: L3 and L4.

Before codegen, t27c checks L3 (ASCII source) only in its lexer, which refuses a non-ASCII identifier. It does not check L4 (a test, an invariant or a bench) at all. t27c test-report counts untested specs, but it generates them. The constitution says L3 covers comments too and that L4 is enforced by the parser (see #8101).

This slice

• By default the core does what t27c does: a non-ASCII name is E_LEX (since P1), and it takes a non-ASCII comment and an untested module, byte-identical with gen-c.

• With strict_laws set, compile also refuses with the new code E_LAW (18):

  • any byte past ASCII, a comment's too (L3);
  • a module with no test, invariant or bench (L4).

  A gate can then enforce the laws without the Rust compiler. The core's own source passes both.

• Core tests for both modes; t27core_ddc.t27 restates strict_laws and is repinned.

The switch is off by default because about 497 of the 1703 specs carry non-ASCII comments and about 134 have no test, invariant or bench, as #8101 records. The core's own tests also compile untested snippets.

Boundary

• specs/compiler/core/t27core_ddc.t27