Optimize decode_gid with flag caching + Add tests and benchmark

[JaskRendix]

PR introduces a performance enhancement to decode_gid, improves test coverage, and adds a lightweight benchmark script to measure decoding speed at scale.

Changes included:

• updated decode_gid to use a flag cache: avoids redundant TileFlags instantiations for repeated GIDs and preserves identical behavior while improving speed
• Added unit tests to cover: cache reuse (identity check), combinations and edge cases for all flag permutations and unpack_gids robustness against empty, malformed, and truncated input
• introduced 2 benchmark scripts

Benchmarks

1. bench_decode_gid.py
   Benchmarks decoding performance with and without caching:

• Measures decoding time across GID sizes (1K to 1M)
• Example results (Python 3.12.3, pygame-ce 2.5.5):

 GID Count |   No Cache |  With Cache |  Speedup
--------------------------------------------------
      1000 | 0.0025s | 0.0027s |    0.92x
      5000 | 0.0059s | 0.0048s |    1.23x
     10000 | 0.0141s | 0.0042s |    3.35x
     50000 | 0.0580s | 0.0207s |    2.80x
    100000 | 0.1213s | 0.0426s |    2.85x
    250000 | 0.2698s | 0.1155s |    2.34x
    500000 | 0.5592s | 0.2202s |    2.54x
   1000000 | 1.0043s | 0.4050s |    2.48x

2. bench_tmx_maps.py
   Benchmarks actual TMX map parsing using files in /apps/data/:

• Measures average load time per map
• Reports tile count, layer count, and tileset usage

  Map Name             | Avg Time (s) | Layers | Tilesets | Tiles
  ----------------------------------------------------------------
  acid0.tmx            |     0.0026   |   14   |     4     |   60
  iso.tmx              |     0.0091   |    6   |     2     |  300
  mapbig1.tmx          |     0.0269   |    4   |     6     |  150
  

[joereynolds]

Out of curiosity did you profile/time this at any point?
Curious how big of an affect this has on loading a map