Log Message:
Update to 0.2.5:

This release not only publishes all database additions that were
contributed in the last half of year, but also introduces several
new improvements in the library itself.

    * As always, many new lenses/camera were added to the database
    * Support for cross-platform compilation using cross-mingw32
    (build instructions in README).
    * Some bugs were fixed in the library, related to memory
    allocation. Now lensfun should be valgrind-clean.
    * Fixed a nasty bug in lfDatabase::FindLenses() that would
    always use fuzzy search (even if you would ask for a 100%
    match). Your applications may chease to "find" lenses if they
    relied on this bug. Just add the LF_SEARCH_LOOSE flag as the
    last parameter to return the old behaviour.
    * Fixed another bug in search logic that sometimes would mark
    unappropiate results with very high scores (esp. when little
    search keywords are given). Now search should work much better.
    * New TCA correction algorithm: LF_TCA_MODEL_POLY3 ("poly3" in
    database). This is almost the algorithm used by Hugin except
    the highest-order power (r^4). This should allow users to use
    tca_correct to compute TCA parameters.
    * Fixed the LF_TCA_MODEL_LINEAR - it used 1/k where it was
    expected to use k. This is the 'v' parameter of Hugin TCA model,
    and previously users had to invert v to get k; now they can
    just use v.
    * New lens calibration tutorial, using Hugin.
    * Some SSE & SSE2 support from Klaus Post with runtime CPU
    features detection.