⚡ Bolt: optimize ZRLE decoding and CopyRect performance - #38

This commit introduces several performance optimizations for X11/VNC rendering:

1.  **ZRLE Decoder Optimizations:**
    *   Promoted `ZInput` to a persistent member of `ZrleDecoder` to reuse its 16KB decompression buffer across rectangles, significantly reducing GC pressure.
    *   Inlined the `avail == 0` check in `ZInput.readByte()` to minimize function call overhead in the hot path.
    *   Implemented strength reduction in `decodeTile` and all subencoding decoders (Raw, Solid, Packed Palette, RLE, Palette RLE) by replacing repeated multiplications and division/modulo operations with incremental row and column counters.
    *   Optimized `ZInput.readCpixel()` to perform direct 3-byte component reads from the internal buffer when available.

2.  **VncClient Optimizations:**
    *   Refactored `ENC_RAW` and `ENC_COPY_RECT` rendering loops to use incremental `rowBase`, `srcRowBase`, and `dstRowBase` pointers, eliminating redundant multiplications per row.
    *   Ensured correct iteration direction in `ENC_COPY_RECT` for overlapping regions.

These changes target the most CPU-intensive parts of the VNC protocol handling on Android (especially ARM architectures), where integer division and repeated coordinate arithmetic are significant bottlenecks.

Tests:
*   `./gradlew :app:testDebugUnitTest --tests "com.excp.podroid.x11.ZrleDecoderTest"`
*   `./gradlew :app:testDebugUnitTest --tests "com.excp.podroid.x11.VncClientTest"`
*   `./gradlew :app:testDebugUnitTest` (full app module suite)

Verification:
*   Verified that all ZRLE subencodings (Raw, Solid, Packed Palette, RLE, Palette RLE) decode correctly.
*   Verified that CopyRect handles overlapping and non-overlapping rectangles correctly.

Co-authored-by: kimocoder <4252297+kimocoder@users.noreply.github.com>