Запрос на слияние 'submit-with-retry' (#1) из submit-with-retry в master

proposed fixes - query last_check and manual flushing; incremental id 'seq' field with locking; safe f.seek calls added just in case; job error pattern matching

retry queue is used to re-submit jobs in order of their addition (earliest wins)

added retry_queue command to locally store queue of retrying job submits

added submit retrial feature to bypass cloud GPU_LIMIT error

fix: one-line default script in yaml templates to avoid special characters error

fix: make default yaml template script run for finite time (100 iterations)

fix: use YAML literal block scalar for multiline script in yaml templates

fix: add safeguards against infinite loop in `job logs -w` and update yaml default scripts

feat: add `mls configure region` command to print selected region for profile

chore: remove WANDB_API_KEY and update default base image to py3.12-torch2.7.0:0.0.41

perf: migrate all_nodes from ThreadPoolExecutor to multiprocessing.Pool

perf: increase HTTP connection pool size to match thread pool in all_nodes

perf: parallelize get_nodes API calls in all_nodes with ThreadPoolExecutor

fix: left-align all columns in all_nodes table

feat: add job description column to all_nodes output

feat: support local .mls/ config with highest priority and --local flag for configure