Delta compression efficiency improvement for non-existent base

Before this CL, when we encoded a sequence with a non-existent
base, we pretended that the delta was 0, and the first delta was
based on that. However, in a sequence where the deltas are small,
but where the first element is big, that would produce
unnecessarily wide deltas. Therefore, we change the behavior in
cases where the base is non-existent, to encode the first existent
value (if any) as a varint; the delta width may then be smaller.

This CL include two piggy-backed changes:
1. Varint encoding/decoding moved to its own file (and an
   additional flavor added).
2. The unit tests for delta encoding are further parameterized
   with a random seed.

Bug: webrtc:8111
Change-Id: I76fff577c86d019c8334bf74b76bd35db06ff68d
Reviewed-on: https://webrtc-review.googlesource.com/c/107860
Reviewed-by: Björn Terelius <terelius@webrtc.org>
Commit-Queue: Elad Alon <eladalon@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#25395}

logging/rtc_event_log/encoder/blob_encoding.h

@@ -18,8 +18,6 @@
 
 namespace webrtc {
 
-extern const size_t kMaxVarIntLengthBytes;
-
 // Encode/decode a sequence of strings, whose length is not known to be
 // discernable from the blob itself (i.e. without being transmitted OOB),
 // in a way that would allow us to separate them again on the decoding side.