Use precedence constraints to enrich timelag constraints and vice versa

[nhuet]

• Normalization of timelag constraints: with negative offsets it is possible to express the same constraints with time lag in min or max list. Moreover duplicates or implied constraints can exist => we normalize the constraints via get_consolidated_time_lags(). Thus we update cpsat-auto:

  • how it finds equality timelags, and avoid adding two equivalent equality constraints
  • avoid adding start_to_end timelags constraints, already encompassed by end_to_start timelags constraints

• Precedence constraints can be seen as end to start min time lags constraints => we add them to consolidated time lags, and we stop adding precedence constraints in cpsat auto to avoid duplicate constraints. We use that normalization in rcpsp/rcpsp_ms when constructing graph

• Conversely, time lag constraints imply precedence constraints and thus should be taken into account into the precedence graph => get_consolidated_precedence_constraints()

In the meantime

• We fix timelag constraints implementation in rcpsp/rcpsp_ms which were
  modifying the original list of timelags given in constructor.
• We remove relax_the_start_at_end parameter used in rcpsp(_ms) satisfy