t27b: integer shifts << >> (first rejection in 36 specs, +32 greedy)

Problem

t27b (cli/t27b) rejects << and >>. On the t27b lab run of master cfff5e4 (https://t27b-lab-production.up.railway.app/runs/cfff5e45309e7803df22a2886ff791285dbed32a.json), ExprBinary(<< >>) is the first rejecting construct in 36 spec files and appears in 63. After f64 (#6286), a local t27b corpus specs --blockers ranks it first in the greedy unlock order, at +32 files.

Shifts compile natively and agree with the reference path (t27c gen + zig test) on every spec that passes there.

<< and >> on the integer types t27b supports, including an untyped literal on the left (1 << n, the case the stale branch claude/t27b-shift 2e075a4 began on: Lower::decl_int, Lower::pin_shifted_literal), and compound <<= / >>= if the front-end produces them.
>> is arithmetic on signed types and logical on unsigned ones, as in Zig. The shift amount follows Zig's rules exactly as the reference emits them (@intCast/std.math.Log2Int, @shlExact or a plain shift). A shift Zig traps on (the amount is too wide for the type) traps the same way in t27b, or the shape is refused with a named blocker. It is never silently masked.
The encoder tests check LSLV/ASRV/LSRV (and immediate forms if they are emitted) against known machine words. There are differential tests named shift_.... A JIT/evaluator disagreement is a failure, never a skip.
The PR gives before/after t27b corpus specs numbers with 0 mismatches, and every newly passing file is checked against the reference path.

cli/t27b/src/{lower,ir,codegen,a64,eval,jit}.rs, cli/t27b/tests/{differential,encoder,blockers}.rs, one docs/now entry.