raw-

Did some measurements on deflate:
For session recording:
- Current batches: 
    In general: deflate for /s was hovering between 18-25% data size, lz64 ~45-55%
    Packing individual events hovered at around 23-27% of data size.


    /s
    deflate: 0.19086889535248105
    lz64: 0.4443071067359241
    raw:
        deflate: 351015
        lz64: 816548
        normal: 1836024
        requests: 60
    /e
    deflate: 0.24228434504792332
    lz64: 0.6569968051118211
    raw:
        deflate: 15167
        lz64: 41128
        normal: 62600
        requests: 17

    Packing individual session recording events:
        count: 287
        normal: 1756870
        packed: 362231
        ratio: 0.20617974010598394

    
- When turning batching up to every 3 seconds:
    /s
    deflate: 0.22700383133167518
    lz64: 0.5097736813031202
    raw:
        deflate: 243675
        lz64: 547548
        normal: 1072595
        requests: 8
    /e
    deflate: 0.11431390879817181
    lz64: 0.5046224161213254
    raw:
        deflate: 4402
        lz64: 19432
        normal: 38508
        requests: 4

Advantage of individual compression: No need to do any backend changes. Can inflate again in the frontend.


Performance measurements:

500kb (posthog.com full snapshot):
- lzstring: 3.98 ops/s +- 2.8%
- pako.deflate: 16.39ops/s +- 8.56%

3kb request (normal /s events, /e)
- lzstring: 165.39 ops/s +- 16.5%
- pako.deflate: 1942.16 ops/s +- 3.75%

300b request (single /e)
- lzstring: 2830.78 ops/s +- 11.84%
- pako.deflate: 2476.65 ops/s +- 6.8%

Note: For some reason our bundled version of lzstring is much slower than the one on cdnjs - did not dig into why.