Concurrent pounders

[jmlarson1]

To help others orient themselves to this PR:

• We are first considering python
• Let conc be the requested concurrency level
• Before any assumed Ffun changes, this PR mocks batching of evaluations in Pounders
• This mock batching happens by storing a Group number and Xtype for each point
• We require nf_max to be a multiple of the requested concurrency level.
• Grouping can happen in 4 places.
  • The "interpolation step" (Xtype 1)
  • The "criticality step" (Xtype 2)
  • The "TRSP step" (Xtype 3)
  • "Model improving" (Xtype 4)
• For the first two places, there is no attempt to ensure only conc points appear. We assume np_max is chosen to align with conc
• For the "TRSP step", TRSPs are solved for delta, delta(1/2),..., delta(1/2^conc-1). If any TRSP solutions satisfy the conditions for being evaluated, all are evaluated
• For the "Model Improving", first the Mdir and -Mdir directions from formquad are considered. The best (positive or negative) from each direction are considered by comparing against the model. Only the conc best directions will be evaluated. (But fewer than conc could be evaluated.)

[coveralls]

Pull Request Test Coverage Report for Build 10582403916

Details

• 119 of 139 (85.61%) changed or added relevant lines in 6 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage increased (+2.9%) to 100.0%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
pounders/py/pounders.py	103	123	83.74%
<!--	Total:	119	139	85.61%	-->

Totals
Change from base Build 10530521992:	2.9%
Covered Lines:	0
Relevant Lines:	0

---

💛 - Coveralls