better support for MongoDB Atlas with cloud workflows

[I'll pause the MongoDB server in Google Cloud. For Vivarium uses we can resume it when needed, and for Fireworks workflows we can use Atlas servers.]

* The updated borealis-fireworks pip supports MongoDB Atlas servers via `uri_mode` and `=` chars in the host metadata strings, e.g. `host=mongodb+srv://.../ecoli1?retryWrites=true&w=majority`.
* Distinguish Fireworker VM names in case the DB name is not unique (`ecoli1` in this example).
* Allocate more RAM to the Fireworker VMs after Parca crashed once in `cvx`. RAM use must've spiked quickly since GCE Monitoring showed it ≈30%.
  ```
  terminate called after throwing an instance of 'std::bad_alloc'
    what():  std::bad_alloc
  Fatal Python error: Aborted
  ...
  Fatal Python error: Segmentation fault
  ```

requirements.txt

@@ -56,7 +56,7 @@ scipy==1.5.4
 aesara==2.0.8
 biopython==1.77
 bokeh==1.4.0
-borealis-fireworks==0.9.1
+borealis-fireworks==0.11.0
 confluent-kafka==1.4.2
 cvxpy==1.1.10
 Cython==0.29.21

runscripts/cloud/util/workflow.py

@@ -471,8 +471,12 @@ def copy_key(src, key, dest):
 			if val is not None:
 				dest[key] = val
 
+		# Construct a name prefix that distinguishes this group of fireworker
+		# VMs from groups that run workflows in other LaunchPad DBs so VM
+		# creation won't hit a name conflict.
 		db_name = config['name']
-		prefix = 'fireworker-{}'.format(db_name)
+		worker_key = f'{self.owner_id}-{db_name}' if self.owner_id != db_name else db_name
+		prefix = f'fireworker-{worker_key}'
 		options = {
 			'image-family': 'fireworker',
 			'network-interface': 'no-address',  # no External IP

runscripts/cloud/util/workflow_cli.py

@@ -12,9 +12,12 @@
 from runscripts.cloud.util.workflow import (DEFAULT_LPAD_YAML,
 	STORAGE_ROOT_ENV_VAR, Task, Workflow)
 
-#: 'custom-1-5120' selects a GCE N1 VM with 1 CPU and 5 GB RAM.
-#: N1's max RAM per CPU is 6656 MB = 6.5 GB.
-GCE_VM_MACHINE_TYPE = 'custom-1-5120'
+#: custom-1-6656: N1 VM with 1 vCPU + 6.5 GB RAM (N1's max RAM per vCPU).
+#: N2, N2D: Should be faster but they charge for 2+ vCPUs while our firetasks
+#:   peg 1 core at 100% and rarely use multiple cores. Worth testing.
+#: e2-medium: 50% of each of 2 shared vCPUs, 100% in bursts. Lower cost. Does it
+#:   sustain 100% of 1 CPU? No control over processor type. Worth testing.
+GCE_VM_MACHINE_TYPE = 'custom-1-6656'
 
 
 class WorkflowCLI(scriptBase.ScriptBase):