Use `Arc<Object>` and `Arc<Vec<Variable>>`

[bvssvni]

This improves the performance of len(list[i]) significantly when the list array contains large objects. The overall performance improves a little.

• Because Variable::Ref is used to point to the data, it means it often stays unique such that Arc::make_mut can be used with little overhead
• The lifetime checker prevents cyclic references, which for Arc means no memory leaks
• Objects and arrays can be shared across multi-threads

With owned Array and Object:
test tests::bench_add    ... bench:  44,275,036 ns/iter (+/- 2,875,523)
test tests::bench_add_n  ... bench:  16,007,990 ns/iter (+/- 825,439)
test tests::bench_array  ... bench:  18,664,707 ns/iter (+/- 3,856,938)
test tests::bench_call   ... bench:  18,135,642 ns/iter (+/- 4,738,525)
test tests::bench_len    ... bench:  71,760,725 ns/iter (+/- 4,210,363) <---
test tests::bench_main   ... bench:   1,838,898 ns/iter (+/- 78,699)
test tests::bench_n_body ... bench: 210,462,478 ns/iter (+/- 17,002,199)
test tests::bench_object ... bench:  25,621,287 ns/iter (+/- 2,244,473)

With Arc:
test tests::bench_add    ... bench:  40,076,479 ns/iter (+/- 2,221,835)
test tests::bench_add_n  ... bench:  14,650,593 ns/iter (+/- 1,469,739)
test tests::bench_array  ... bench:  17,662,768 ns/iter (+/- 1,859,916)
test tests::bench_call   ... bench:  16,366,031 ns/iter (+/- 1,277,929)
test tests::bench_len    ... bench:   3,036,349 ns/iter (+/- 261,624) <---
test tests::bench_main   ... bench:   1,845,956 ns/iter (+/- 85,599)
test tests::bench_n_body ... bench: 200,154,231 ns/iter (+/- 7,275,241)
test tests::bench_object ... bench:  25,420,239 ns/iter (+/- 3,226,371)