byteAlignBlocks

    pub fn byteAlignBlocks(b: *BitWriter) Writer.Error!void {
        if (b.buffered_n == 0) return;

        // There are two methods to do this:
        // 1. A store block (5 or 6 bytes)
        // 2. Outputting empty 10-bit fixed blocks until aligned
        //
        // Fixed blocks advance the bit alignment by two, and so can only used for even numbers
        // requiring a maximum of four bytes (three blocks = 30 bits) to which is always more
        // efficient than store blocks.
        if (b.buffered_n & 1 == 0) {
            const splat = (8 - @as(u5, b.buffered_n)) >> 1;
            const bits = splat * 10;
            // fixed eos code is 0, so the only bits are for the block header
            const pattern: u32 = BlockHeader.int(.{ .kind = .fixed, .final = false });
            const splatted = ((pattern << 20) | (pattern << 10) | pattern) >> (30 - bits);
            try b.write(splatted, bits);
        } else {
            try b.write(BlockHeader.int(.{ .kind = .stored, .final = false }), 3);
            try b.output.rebase(0, 5);
            b.byteAlign();
            b.output.writeInt(u16, 0x0000, .little) catch unreachable;
            b.output.writeInt(u16, 0xffff, .little) catch unreachable;
        }

        assert(b.buffered_n == 0);
    }