were writing in text-mode in PY3. and failing as encoding error.

receiving partial inputs, needed for other operations.

+ The x2 TCs added just before are now passing.

NOTE dict are not deterministic in <PY3.6.
So this commit would not improve determinism
in those pythons.

+ build: add `boltons` dependency for ndexedSet.
+ doc: mark all set usage if affect determinism.
+ e.g. see reproducibility problem in yahoo#14.

…uts asked

needed to refactor the new pruning algorithm.

- expected 2 TCs fail for yet-to-be-solved yahoo#24 & yahoo#25 bugs.

the later described in yahoo#21.

override intermediate data.

More changes only for newer pruning TCs:
+ refact(test): rename graph-->netop vars for results of compose(),
  to avoid of `graph.net.graph`.
+ Explain failure modes in v1.2.4 & this merged branch (yahoo#19 + yahoo#23).

not after compose().

+ All TCs pass ok.
+ NOTE this is not yet what is described in yahoo#21.

to pass +TC checking DeleteInst vary when inputs change.

- x4 TCs still failing, and need revamp of dag-solution.

+ Read the next doc-only commit to understand changes.
+ Renamed:
  + net.steps --> net.execution_plan.
  + (old)compile() --> _build_execution_plan()
  + _find_necessary_steps() --> (new)compile() + _solve_dag()
    compile() became the master function invoking _solve_dag &
    _build-execution_plan(), and do the caching.
+ refact(compute()): extract common tasks from sequential/parallel.
+ refact show_layers() to allow full-print, geting also string
  (not just printing), and using custom classes for representation.
+ Raise AssertionError when invalid class in plan.
  it's a logic error, not a language type-error.

Probaly unreported bug in v1.2.4 for '_neccessary_steps_cache`.

+ Pruning behaves correctly also when outputs given;
  this happens by breaking incoming provide-links
  to any given intermedediate inputs.
+ Unsatisfied detection now includes those without outputs
  due to broken links (above).
+ Remove some uneeded "glue" from unsatisfied-detection code,
  leftover from previous compile() refactoring.
+ Renamed satisfiable --> satisfied.
+ Improved unknown output requested raise-message.
+ x3 TCs PASS, x1 in yahoo#24 and the first x2 in yahoo#25.
- 1x TCs in yahoo#25 still FAIL, and need "Pinning" of given-inputs
  (the operation MUST and MUST NOT run in these cases).

…must run

- WIP: PARALLEL execution not adding PINS!
+ Insert "PinInstructions" in the execution-plan to avoid overwritting.
+ Add `_overwrite_collector` in `compose()` to collect re-calculated values.
+ FIX the last TC in yahoo#25.

- STILL buggy PIN on PARALLEL, 2 DISABLED TCs FAIL:
  - test_pruning_with_given_intermediate_and_asked_out()
  - test_unsatisfied_operations_same_out()
+ move check if value in asked outputs before cache-evicting it
  in build-execution-plan time - compute methods
  don't need outputs anymore.
+ test: speed up parallel/multihtread TCs
  by reducing delays & repetitions.
+ refact: network rightfully adopted stray functions
  for parallel processing - they all worke on the net.graph,
+ upd: networkx api by indexing on `dag.nodes` views.
+ enh: add log message when deleting in parallel
  (in par with sequential code).
+ refact: var-renames, if-then-else simplifications, pythonisms.
+ doc: A lot!

- WIP: x4 TCs FAIL and still not discovered th bug :-(
+ BUT ALL+AUGMENTED PARALLEL TCs pass
  (yahoo#26 were failing some)
+ refact: net stores also `pruned_dag` (not only `steps`).
+ refact: _solve_dag() --> _prune_dag().
+ doc: +a lot.
+ TODO: store pruned_dag in own ExePlan class.

... bugged in the opening commit d403783 of this PR, and
discovered 68(!) commits later, and all that time had to live
with x4 broken TCs with asked-outputs.

+ Partial fix deterministic results (yahoo#22-2.4.3i).

bc subgraph was taken on plain string outputs.

+ minor upd err-msg terminology.

due to bad node check, evicting parallels it nevered kicked in.