[tx] Compress flags on insertion and have SQLite expand internally

[ncalexan]

In the Clojure implementation, we do a lot of "wide inserts" where we insert 10 columns, 4 of which are bit flags. In the following, index_avet, index_vaet, index_fulltext, and unique_value are bit flags:

INSERT INTO datoms (e, a, v, value_type_tag, tx, index_avet, index_vaet, index_fulltext, unique_value) VALUES (...)

After #214, we'll be inserting into a "to search" table and populating a "search results" table. We could write a single bit field into the "to search" table, coalescing 4 bit flags into 1 bit field, and then have SQLite populate the "search results" table with bit flags. This would reduce the total number of insert statements we have to execute, which should speed everything up.

This ticket tracks trying this and seeing if it's a perf win: it's possible SQLite does this very slowly, or the overhead is negligible, or ...

[ncalexan]

Concretely, this would mean:

1. changing the temporary exact_searches, inexact_searches, and search_results schemas to have flags INTEGER NOT NULL
2. updating the INSERT INTO *_searches statements to bitwise-or the relevant flags together;
3. updating the INSERT INTO search_results statements to copy the flags from the search tables;
4. updating the INSERT INTO datoms statements to expand the flags into the index fields using SQLite's bitwise arithmetic operators (see https://www.sqlite.org/lang_expr.html).
5. performance testing!

[ncalexan]

@jsantell this might be a [good next bug], and a good way to dig into the transactor internals a little.

[ncalexan]

I realize that we can drop tx entirely in the "to search" tables. Further still, we can drop value_type_tag, since we always compare with l.a = r.a, which guarantees that the corresponding l.value_type_tag = r.value_type_tag. This will help us get datoms on to disk still faster!