Add memcmp short-circuit to ST_Equals

[dbaston]

As described in pull request 44 (​https://github.com/postgis/postgis/pull/44), this can be very beneficial for certain types of queries.

Patch at ​https://patch-diff.githubusercontent.com/raw/postgis/postgis/pull/44.patch

[strk]

r13864, thanks !

[robe]

Not a huge deal, but wouldn't this cause a regression for invalid geometries? I think before even if they were mem equal, they were invalid because of not well-defined space.

[strk]

This change might have broken interruption test, see #3226

[strk]

robe2 about regression on invalid geometries, what would you expect from ST_Equals in that case ? A false, a null or an exception ? None of the relate operations are _guaranteed_ to return a specific value on invalid input, as it would require checking for validity and it's expensive.

[strk]

Closing, after confirming that the interrupt tests only failed because the ST_Equals call performed by the test is now faster than the timeout. Case fixed by changing the testcase to make not as easy to compute equals :)