Neurodiffeq

- new condition: `DoubleEndedBVP1D`
- more robust `MonitorCallback`
- hierarchy issue fix in `neurodiffeq.monitors`

1. Refactored all `solve*` functions to `Solver` classes
2. Renamed some arguments for consistency
3. Implemented the `Swish` activation
4. Cleaned & updated the documentation
5. Miscellaneous fixes & enhancement

This release focuses on new APIs and backward compatibility.
- Addressed argument naming issues 70
- Addressed ambiguity in FCNN's constructor 71
- Addressed a potential bug in the `diff` operator
- Included new test cases for warnings and errors

- all conditions are moved to conditions.py, with custom test cases implemented.
- better styles for sphinx docs & misc docs fixes
- fixed travis-ci bug and re-enabled code coverage
- diff will check tensor shape by default
- Basic Dash documentation support
- use safe_diff internally and in test cases (except for test_temporal.py)
- Functionality to set default dtype and device for tensors
- Use 64-bit float (a.k.a. double) by default.
- diff in v0.2.0 checks shape by default.