Fix memory64 offset truncation in the interp istream

[aizu-m]

Running a memory64 module under wasm-interp, i64.load offset=0x1_0000_0000 from address 0 returned 0 instead of trapping. A large dynamic address (5 GiB) does trap, so the static offset was the suspect.

1. OnLoadExpr hands the memarg offset over as an Address (u64), but istream_.Emit(opcode, memidx, offset) resolved to Emit(Opcode, u32, u32) and narrowed it to 32 bits.
2. For an i64-indexed memory a static offset of 2^32 or more is then lost, so an access whose effective address is out of bounds wraps to a small in-range offset and reads or writes a valid location rather than trapping.

Carried the offset as u64 the whole way: added imm_index_offset/imm_index_offset_lane immediates and Emit(Opcode, u32, u64) overloads, widened the four Imm_Index_Offset_* decode groups, the load/store/atomic/simd-lane handlers and the disassembler. 32-bit memories are untouched since the offset always fits u32. Regression test added under test/interp.

[zherczeg]

I would move the lane to first member (or after memIndex) to reduce the structure size increase.
u32+u64+u8 = 24 bytes, while u8+u32+u64 = 16 bytes due to padding.