fix: Workaround for mmap crash under Emscripten

[georgestagg]

When running Polars in Pyodide, using read_csv() (and possibly other functions reading from file) causes a crash during mmap -- a huge chunk of memory is allocated as the length of the file is incorrectly provided to mmap by the file's metadata.

This PR implements a workaround that applies only when running under Emscripten so that the file length is derived by seeking to the end of the file, and setting the mmap size explicitly in the options before mapping.

With this, things work under Pyodide again:

Emscripten is currently working on reimplementing the virtual filesystem from scratch (WasmFS). Once that work has been merged and hits Pyodide, it's likely that we can remove this workaround.

cc @ritchie46

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 78.99%. Comparing base (2ae01fb) to head (199b748).
Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #20418   +/-   ##
=======================================
  Coverage   78.98%   78.99%           
=======================================
  Files        1562     1562           
  Lines      220103   220103           
  Branches     2486     2486           
=======================================
+ Hits       173855   173863    +8     
+ Misses      45675    45667    -8     
  Partials      573      573           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ritchie46]

Nice. I think I will follow up ensuring that all our mmap's go through a single utility function.

Note; the flaky test is unrelated.