vsicurl directory parsing slow with large directories

[warmerdam]

Parsing large directory listings via the vsicurl dirver seems to be painfully slow.

eg.

​http://hyperquad.telascience.org/naipsource/Texas/20100801/naip10_1m_2597_03_2_20100424.tif

The problem seems to be ordern(n-squared) behavior in the directory parsing. ie. use of strstr() to find the next of something which can be expensive in a long document if it is not found for each entry.

I will dig into this a bit.

[warmerdam]

I have correct the worst offending part of the code in trunk (r22728). There is still an n-squared problem with adding the filenames to the filename list but I'm addressing that a bit distinctly since my approach will be more complicated than is suitable to back port into 1.8.

[warmerdam]

I also patched things to use a more efficient allocation strategy for the string list of filenames (r22731). Due to use of CPLStringList this is not suitable to back port.