Add option to skip model statistics collection before solve (context.solver.collect_model_statistics)

[a-safari]

Summary

This PR adds an opt-in configuration flag, context.solver.collect_model_statistics,
that allows callers to skip the model statistics / KPI metadata collection that
CplexLocalSolveEnv.before_solve() currently performs unconditionally before every
local CPLEX solve.

Motivation

Profiling a MIP-heavy application that calls Model.solve() many times per run
(hundreds of solves) showed that Model.statistics (and the _make_new_stats()
call it triggers) was one of the largest contributors to wall-clock time outside
of CPLEX itself.

Looking at CplexLocalSolveEnv.before_solve(), this cost comes from an
unconditional block that:

1. Computes mdl.statistics (iterating all variables and constraints to count
   binary/integer/continuous/semi-continuous/semi-integer variables and all
   constraint types).
2. Builds a KPI metadata dictionary (make_new_kpis_dict(...)) from those
   statistics.
3. Publishes that metadata via the_env.notify_start_solve(kpis) and,
   optionally, the_env.update_solve_details(...).

Applications that do not consume this solve-start metadata pay this
cost on every single solve with no benefit.

There is currently no way to opt out of this calculation through the public
Context API — context.solver.auto_publish.solve_details only controls
whether the already computed metadata is published via
update_solve_details(...); it does not prevent mdl.statistics from being
computed in the first place.

Changes

• docplex/mp/context.py

  • Added a new SolverContext attribute, collect_model_statistics, defaulting
    to True to preserve existing behavior for all current users.
  • Documented the new attribute alongside the other solver.* settings.

• docplex/mp/solve_env.py

  • Wrapped the statistics/KPI construction block inside
    CplexLocalSolveEnv.before_solve() in an
    if context.solver.collect_model_statistics: check.
  • When the flag is False, the method now calls
    the_env.notify_start_solve({}) instead, preserving the solve-start
    notification lifecycle without computing statistics or KPI metadata.
  • All other behavior in before_solve() is unchanged.
  • after_solve() is unmodified.

Backward compatibility

• Default value is True, so behavior is unchanged unless a caller explicitly
  sets context.solver.collect_model_statistics = False.
• No public API is removed or renamed.

How to use

from docplex.mp.model import Model

model = Model(name="my_model")
model.context.solver.collect_model_statistics = False