use optional index in multivalued index

[PSeitz]

For mostly empty multivalued indices there was a large overhead during
creation when iterating all docids. This is alleviated by placing an
optional index in the multivalued index to mark documents that have values.

There's some performance overhead when accessing values in a multivalued
index. The accessing cost is now optional index + multivalue index. The
sparse codec performs relatively bad with the binary_search when accessing
data. This is reflected in the benchmarks below.

This changes the format of columnar to v2, but code is added to handle the v1
formats. This is done by passing down the global version flag to the index. We could have the version in each column, this would make version handling slightly less awkward. But it adds some extra bytes, which may add up, if there are many columns.

     Running benches/bench_access.rs (/home/pascal/Development/tantivy/optional_multivalues/target/release/deps/bench_access-ea323c028db88db4)
multi sparse 1/13
access_values_for_doc        Avg: 42.8946ms (+241.80%)    Median: 42.8869ms (+244.10%)    [42.7484ms .. 43.1074ms]
access_first_vals            Avg: 42.8022ms (+421.93%)    Median: 42.7553ms (+439.84%)    [42.6794ms .. 43.7404ms]
multi 2x
access_values_for_doc        Avg: 31.1244ms (+24.17%)    Median: 30.8339ms (+23.46%)    [30.7192ms .. 33.6059ms]
access_first_vals            Avg: 24.3070ms (+70.92%)    Median: 24.0966ms (+70.18%)    [23.9328ms .. 26.4851ms]
sparse 1/13
access_values_for_doc        Avg: 42.2490ms (+0.61%)    Median: 42.2346ms (+2.28%)    [41.8988ms .. 43.7821ms]
access_first_vals            Avg: 43.6272ms (+0.23%)    Median: 43.6197ms (+1.78%)    [43.4920ms .. 43.9009ms]
dense 1/12
access_values_for_doc        Avg: 8.6184ms (+23.18%)    Median: 8.6126ms (+23.78%)    [8.5843ms .. 8.7527ms]
access_first_vals            Avg: 6.8112ms (+4.47%)     Median: 6.8002ms (+4.55%)     [6.7887ms .. 6.8991ms]
full
access_values_for_doc        Avg: 9.4073ms (-5.09%)    Median: 9.4023ms (-2.23%)    [9.3694ms .. 9.4568ms]
access_first_vals            Avg: 4.9531ms (+6.24%)    Median: 4.9502ms (+7.85%)    [4.9423ms .. 4.9718ms]

     Running benches/bench_merge.rs (/home/pascal/Development/tantivy/optional_multivalues/target/release/deps/bench_merge-475697dfceb3639f)
merge_multi 2x_and_multi 2x                          Avg: 20.2280ms (+34.33%)    Median: 20.1829ms (+35.33%)    [19.9933ms .. 20.8806ms]
merge_multi sparse 1/13_and_multi sparse 1/13        Avg: 0.8961ms (-78.04%)     Median: 0.8943ms (-77.61%)     [0.8899ms .. 0.9272ms]
merge_dense 1/12_and_dense 1/12                      Avg: 0.6619ms (-1.26%)      Median: 0.6616ms (+2.20%)      [0.6473ms .. 0.6837ms]
merge_sparse 1/13_and_sparse 1/13                    Avg: 0.5508ms (-0.85%)      Median: 0.5508ms (+2.80%)      [0.5420ms .. 0.5634ms]
merge_sparse 1/13_and_dense 1/12                     Avg: 0.6046ms (-4.64%)      Median: 0.6038ms (+2.80%)      [0.5939ms .. 0.6296ms]
merge_multi sparse 1/13_and_dense 1/12               Avg: 0.9111ms (-83.48%)     Median: 0.9063ms (-83.50%)     [0.9047ms .. 0.9663ms]
merge_multi sparse 1/13_and_sparse 1/13              Avg: 0.8451ms (-89.49%)     Median: 0.8428ms (-89.43%)     [0.8411ms .. 0.8563ms]
merge_multi 2x_and_dense 1/12                        Avg: 10.6624ms (-4.82%)     Median: 10.6568ms (-4.49%)     [10.5738ms .. 10.8353ms]
merge_multi 2x_and_sparse 1/13                       Avg: 10.6336ms (-22.95%)    Median: 10.5925ms (-22.33%)    [10.5149ms .. 11.5657ms]

Based on: #2429

Closes #2431