zstd: Improve better compression (#364)

Try to find a better match by searching for a long match
at the end of the current best match

Before/after pairs.. Speed comparison not reliable, different Go versions.

```
silesia.tar zskp    3   211947520   65177448    1899    106.44
silesia.tar zskp    3   211947520   64595893    2007    100.68

gob-stream  zskp    3   1911399616  185792019   9324    195.48
gob-stream  zskp    3   1911399616  175034659   9636    189.17

enwik9  zskp    3   1000000000  294540704   11725   81.34
enwik9  zskp    3   1000000000  292243069   12162   78.41

github-june-2days-2019.json zskp    3   6273951764  537511906   29252   204.54
github-june-2days-2019.json zskp    3   6273951764  524340691   34043   175.75

rawstudio-mint14.tar    zskp    3   8558382592  3224594213  71751   113.75
rawstudio-mint14.tar    zskp    3   8558382592  3158085214  77675   105.08

nyc-taxi-data-10M.csv   zskp    3   3325605752  538490114   25683   123.49
nyc-taxi-data-10M.csv   zskp    3   3325605752  530289687   25239   125.66
```

zstd/enc_better.go

@@ -412,8 +412,41 @@ encodeLoop:
 			cv = load6432(src, s)
 		}
 
-		// A 4-byte match has been found. Update recent offsets.
-		// We'll later see if more than 4 bytes.
+		// Try to find a better match by searching for a long match at the end of the current best match
+		if true && s+matched < sLimit {
+			nextHashL := hash8(load6432(src, s+matched), betterLongTableBits)
+			cv := load3232(src, s)
+			candidateL := e.longTable[nextHashL]
+			coffsetL := candidateL.offset - e.cur - matched
+			if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) {
+				// Found a long match, at least 4 bytes.
+				matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4
+				if matchedNext > matched {
+					t = coffsetL
+					matched = matchedNext
+					if debugMatches {
+						println("long match at end-of-match")
+					}
+				}
+			}
+
+			// Check prev long...
+			if true {
+				coffsetL = candidateL.prev - e.cur - matched
+				if coffsetL >= 0 && coffsetL < s && s-coffsetL < e.maxMatchOff && cv == load3232(src, coffsetL) {
+					// Found a long match, at least 4 bytes.
+					matchedNext := e.matchlen(s+4, coffsetL+4, src) + 4
+					if matchedNext > matched {
+						t = coffsetL
+						matched = matchedNext
+						if debugMatches {
+							println("prev long match at end-of-match")
+						}
+					}
+				}
+			}
+		}
+		// A match has been found. Update recent offsets.
 		offset2 = offset1
 		offset1 = s - t