Meadowrun

Improvements:
- Now compatible with any cloudpickle version
- Switch from conda to mamba for container environments (we were already using mamba for directly-built environments)
- Add support for the `conda list --explicit` format (in addition to the conda env export format). Make this format the default for `mirror_local` from a conda environment

**Full Changelog**: https://github.com/meadowdata/meadowrun/compare/v0.2.15...v0.2.16

Improvements and bug fixes:
- Call os.chmod in meadowrun-manage-ec2 get-ssh-key so that the keys are immediately ready to use on Linux
- Give meadowrun-managed EC2 instances names
- For mdr-reusable pods in Kubernetes, use an alphanumeric string instead of a uuid to make the names unique
- Rename the meadowrun_EC2_alloc tag to just "meadowrun"
- Handle the case in mirror_local where there is a .venv folder on your current working directory
- Support for editable install in pip requirements files in git_repo

**Full Changelog**: https://github.com/meadowdata/meadowrun/compare/v0.2.14...v0.2.15

Public API changes:
- For all of the run_* functions, deployment=None now implies Deployment.mirror_local. Previously, this would mean PreinstalledInterpreter(MEADOWRUN_INTERPRETER)
- Kubernetes(resuable_pods=True) is now the default (previously was False)
- Add Deployment.preinstalled_interpreter
- Remove LocalPipInterpreter(python_version=) parameter, just detect it from the interpreter
- In mirror_local, renamed additional_python_paths to additional_sys_paths, add a parameter include_sys_path_contents, and rename working_directory_globs to globs. globs can now specify files in sys.path (previously was limited to files in the current working directory)
- Add PipRequirementsString as a way to specify an interpreter

New features:
- Add support for using Google Cloud Storage with Kubernetes, means that GKE is supported in a first-class way
- Azure now uses incremental code upload
- Kubernetes has a pod_customization
- extra-index-url/index-url config in pip.conf will be synced automatically by mirror_local
- Include .so files in mirror_local by default

Bug fixes:
- Single use pods were not working
- Reusable pods weren't being killed correctly
- Add support for streaming logs with reusable pods
- Add a Resources(ephemeral_storage_gb) parameter that works only for Kubernetes for now
- Allow functions in run_map to call asyncio.run
- mirror_local can now be reused as long as its with the same host
- Provide a default None value for Kubernetes.storage_spec = None
- If a conda and pip environment were both active, mirror_local would always pick the conda environment even if the pip one was "higher priority"
- Don't fail if a file can't be read for zipping in mirror_local
- Allow using different versions of python with the same pip/poetry environment
-

**Full Changelog**: https://github.com/meadowdata/meadowrun/compare/v0.2.11...v0.2.13

Improvements/new features:
- Code is now uploaded in chunks, so a small change won't result in re-uploading the entire codebase
- Pip packages from Google Cloud Artifact Repository can be referenced using keyring-based authentication. Useful for running on GKE.

Bug fixes:
- Fixes an issue in run_map where any task results that were not built-in types would result in an error

**Full Changelog**: https://github.com/meadowdata/meadowrun/compare/v0.2.10...v0.2.11

New features:
- Complete v1 of reusable pods implementation for Kubernetes

Bug fixes and improvements:
- Better support for run_map tasks that exit unexpectedly (e.g. segfault)
- Better cleanup of S3/object storage files
- Compress uploaded code
- Long run_map arguments/results were being truncated

**Full Changelog**: https://github.com/meadowdata/meadowrun/compare/v0.2.9...v0.2.10

Features:
- Add an `edit-management-lambda-config` command that allows configuring the `TERMINATE_INSTANCES_IF_IDLE_FOR_SECS` parameter
- Add initial support for a `retry_with_more_memory` parameter in `run_map`
- Add experimental support for `reusable_pods` parameter in Kubernetes

Bug fixes:
- Exclude p2 instances from allocation until we can implement support for using an old version of CUDA (the GPUs on these instance types require an old version of CUDA)

**Full Changelog**: https://github.com/meadowdata/meadowrun/compare/v0.2.8...v0.2.9