dsc#92: give bytes a member catalog starting at .length

[samifouad]

Closes dsc#92. Tracker: dsc#66. Related: dsc#89, dsc#88, dsc#90, deka#756.

What

bytes gets a member catalog with exactly one member: .length. This is the entire change to the language surface.

const r = unsafe<bytes> { new Uint8Array([65, 66, 67]) }
echo(match (r) {
  Ok(v) => `${v.length}:${v[0]}:${v[2]}`,   // "3:65:67"
  Err(e) => `err: ${e.message}`,
})

RFD 15 justification, member by member

RFD 15 (dekaruntime/rfd#15) specifies the bytes operations as len, slice, concat, to_hex, from_hex, to_base64, from_base64, from_string, and a fallible to_string. None of the operation names are specified as member-method shape — the RFD never commits to member access vs free functions, and how each lands is deka#756's call. This PR adds only what is unambiguous and provably missing:

• .length — added. This is RFD 15's len operation. It is surfaced as the .length property to match the convention string.length and array.length already establish (primitive_member in crates/deka_syntax/src/typeck/expr.rs), not as a len member that no other primitive has. The bytes package already exposes free-function len (tests/testsuite/packages/bytes_from_string_len), so the operation exists on both shapes the RFD's naming leaves open.
• slice, concat — deliberately not added. Both are in the RFD's operation list, but exposing them as members now would pre-empt deka#756's design decision (member vs free function, and for slice, whether bounds behaviour gets DekaScript semantics rather than JS's). The unknown-member diagnostic names what exists, so a user who reaches for b.slice gets `bytes` has no field `slice` (available: `length`) at check time, not a silent gap.
• to_hex/from_hex/to_base64/from_base64/from_string/to_string — not added. These are conversions with open design questions in the RFD itself (fallible shape: Option vs Result is RFD 15 open question 1). Landing them here would ship ahead of the design; that is deka#756.

Nothing here contradicts or exceeds RFD 15: the one member added is the one operation the RFD guarantees, under the spelling the rest of the language's primitives use.

The catalog is closed

resolve_primitive_field gains a dedicated bytes arm, so a nonexistent member is a check-time error with a span and a message naming the one available member (the JsError diagnostic pattern). Unit tests pin both directions: bytes.length is a number property (and rejects flow into string), and b.slice errors at line 2, column 10. Indexing (b[0], dsc#88) is untouched.

Emission cost: zero

bytes is a Uint8Array at runtime and the emitter's FieldAccess arm is transparent, so v.length compiles to exactly v.length — what you would hand-write. A transpile test (bytes_length_transpiles_verbatim in crates/cli/tests/transpile.rs) pins that: the emitted JS contains b.length and no __deka helper.

Fixtures

• tests/testsuite/data_types/bytes_length — executed (stage run), exact stdout 3:65:67, exercising the dsc#89 regression case directly: ${v.length} inside template interpolation on a narrowed bytes.
• tests/testsuite/types/bytes_unknown_member — negative fixture, unknown member rejected with the expected diagnostic.
• Once this ships and deka bumps its scripts/dsc-version pin past it, the two ratcheted lines in deka's expected-failures.txt (unsafe-textencoder-encode, unsafe-uint8array-literal) can come out per their stated removal condition.

Test output

cargo test --workspace — all suites green (incl. the two new checker tests and the new transpile test; the only warnings are two pre-existing unused-variable ones also present on main).

test result: ok. 26 passed ... (all 28 suites ok, 0 failed)

Testsuite against the pinned runtime (0.46.0) plus this branch's dsc:

native CLI: deka 0.46.0 | fixtures: 874  jobs: 8
 Passed: 719 | Failed: 0 | Known: 1 | Skipped: 154 | Total: 874

Both new fixtures verified individually: bytes_length passes at stage run with stdout 3:65:67; bytes_unknown_member fails check with [transpile] 2:10: byteshas no fieldslice(available:length).

-kimi