`meta_log` & `hyper_log` class instances w. functionalities (`filter`, `plot`)

[RobertTLange]

Currently we have to call separate functions to filter the meta_log for specific runs or the hyper_log for variables:

from mle_toolbox.utils import load_result_logs, subselect_meta_log, subselect_hyper_log
meta_log, hyper_log = load_result_logs("/some_results_dir/")

# Filter hyper log for fixed variable
subselect_hyper_log(hyper_log, "inner_lrate", 0.001)
subselect_meta_log(meta_log, ["b_1_eval_0", "b_1_eval_1"])

Instead I want hyper_log and meta_log to be classes with a filter method. E.g.

hyper_log.filter("inner_lrate", 0.001)
meta_log.filter(["b_1_eval_0", "b_1_eval_1"])

Additional features could include:

• [x] Plotting 1d and 2d directly from log.
• [x] Hypothesis testing one run vs another.
• [ ] Summary stats top-k performing models.

Also move load_result_logs import to base of mle_toolbox for easier workflow.