GoldenFloat course specs: phi_ratio, phi_split_optimality, radix_economy discard 331 tokens on master and assert false things #7804

Problem

Three specs the t27.ai GoldenFloat course teaches from do not survive the site's compiler on master, and several of their tests assert false things:

Examples of false claims: phi_ratio's GF24/GF32 rows say the phi split is 6/17 and 8/23 and does not match (it is 9/14 and 12/19, and matches); radix_economy says 27 balanced trits need 43 bits (they need 42); ln with 4-5 series terms misses its own test tolerances.

The corrected versions exist only in gHashTag/trinity's vendored mirror (apps/website/public/t27/files), so every catalog refresh reverts them to master and the course loses its lessons (gHashTag/trinity#1540, fixed on the site side by trinity#1539, which keeps the course's copy until master has a clean one).

Solution

Take the corrections here: PR #7496 carries them byte-identically to trinity's copy (it also removes the cannot assign to immutable warnings from gf8, gf12, gf20, gf24, gf32 with let mut).

Acceptance

• The three specs consume every token under t27c parse-complete and compile clean on all seven backends in the site's wasm compiler.
• Seals for the eight touched specs refreshed (t27c seal <spec> --save).

User Scenarios

• Given the three GoldenFloat course specs on master, when t27c parse-complete runs on each, then all tokens are consumed (zero discarded) and each spec compiles cleanly on all seven backends in the site's wasm compiler.
• Given the corrected spec contents from PR spec(numeric,math): the GoldenFloat lesson specs compile clean and say only true things #7496, when the specs are updated in this repository, then the false assertions (phi split values, radix economy bit count, ln series tolerances) are replaced with correct values and all internal tests pass.
• Given the eight specs referenced in the acceptance criteria, when t27c seal <spec> --save is run for each, then the seal files are refreshed and match the corrected behavior.

Requirements

• FR-001: The three cited specs MUST consume every token under t27c parse-complete with zero top-level tokens discarded.
• FR-002: The three cited specs MUST compile cleanly on all seven backends in the site's wasm compiler.
• FR-003: phi_ratio.t27 MUST assert correct phi split values (GF24: 9/14, GF32: 12/19) and those splits MUST match.
• FR-004: radix_economy.t27 MUST assert that 27 balanced trits need 42 bits (not 43).
• FR-005: ln series tests with 4-5 terms MUST meet their stated tolerances.
• FR-006: The let mut bindings for gf8, gf12, gf20, gf24, gf32 MUST be applied to eliminate "cannot assign to immutable" warnings.
• FR-007: Seals for all eight touched specs MUST be refreshed via t27c seal <spec> --save.

Success Criteria

• t27c parse-complete specs/numeric/phi_ratio.t27 exits 0 with 0 tokens discarded
• t27c parse-complete specs/math/phi_split_optimality.t27 exits 0 with 0 tokens discarded
• t27c parse-complete specs/math/radix_economy.t27 exits 0 with 0 tokens discarded
• All three specs compile clean on all seven wasm backends (site compiler test passes)
• t27c seal specs/numeric/phi_ratio.t27 --save exits 0 and updates seal
• t27c seal specs/math/phi_split_optimality.t27 --save exits 0 and updates seal
• t27c seal specs/math/radix_economy.t27 --save exits 0 and updates seal
• 5 additional spec seals refreshed (gf8, gf12, gf20, gf24, gf32) via t27c seal --save each exiting 0

Boundary

• specs/numeric/phi_ratio.t27
• specs/math/phi_split_optimality.t27
• specs/math/radix_economy.t27