Optimise read planning.

Previously we had to do do a linear search through the list of
segment reference to find which file to read a given entry from.

This commit introduced a new module `ra_lol` which is a
"list of list" type of ordered data structure that replaces
the use of a plain list.

ra_log keeps data in a list of 64 entry sized "rows" which
subtantially reduces the amount of work needed to find a
given segment ref as we first scan each row to find the
row with the segref in then we can the fixed size row.

Even a server with 4096 segment only need to scan at most
128 to find any given entry.

This commit also fixes a few issues in segref compaction
and changes the type of the segment ref to a {ra_range:range(), filename()}
to work better with the ra_range module.

src/ra_log.erl

@@ -64,8 +64,7 @@
 -define(WAL_RESEND_TIMEOUT, 5000).
 
 -type ra_meta_key() :: atom().
--type segment_ref() :: {From :: ra_index(), To :: ra_index(),
-                        File :: file:filename_all()}.
+-type segment_ref() :: {ra_range:range(), File :: file:filename_all()}.
 -type event_body() :: {written, ra_term(), ra:range()} |
                       {segments, [{ets:tid(), ra:range()}], [segment_ref()]} |
                       {resend_write, ra_index()} |
@@ -279,7 +278,7 @@ init(#{uid := UId,
     LastSegRefIdx = case SegRefs of
                         [] ->
                             -1;
-                        [{_, L, _} | _] ->
+                        [{{_, L}, _} | _] ->
                             L
                     end,
     LastWrittenIdx = case ra_log_wal:last_writer_seq(Wal, UId) of
@@ -969,7 +968,7 @@ should_snapshot(snapshot, Idx,
     % We should take a snapshot if the new snapshot index would allow us
     % to discard any segments or if the we've handled enough commands
     % since the last snapshot.
-    CanFreeSegments = lists:any(fun({_, To, _}) -> To =< Idx end,
+    CanFreeSegments = lists:any(fun({{_, To}, _}) -> To =< Idx end,
                                 ra_log_reader:segment_refs(Reader)),
     CanFreeSegments orelse Idx > SnapLimit;
 should_snapshot(checkpoint, Idx,
@@ -1331,7 +1330,7 @@ recover_ranges(UId, MtRange, SegWriter) ->
                                 [SegRef | Acc]
                         end
                 end, [], SegFiles),
-    SegRanges = [{F, L} || {F, L, _} <- SegRefs],
+    SegRanges = [Range || {Range, _} <- SegRefs],
     Ranges = [MtRange | SegRanges],
     {pick_range(Ranges, undefined), SegRefs}.