Fix subtraction of wait_until_time not accounting for NEVER

[SheaFixstars]

The subtraction of the _lf_action_delay_table entry from the wait_until_time value is causing issues.

When the after keyword is used the delay is set accordingly. But when the after keyword is NOT used then the _lf_action_delay_table entry gets set to NEVER that maps to LLONG_MIN.

According to the inline documentation located shown below and located at https://github.com/lf-lang/lingua-franca/blob/0bb4bf973f24c3f484e42fa0816af3315ff6828d/core/src/main/java/org/lflang/federated/extensions/CExtensionUtils.java#L163

/**
   * Given a connection 'delay' expression, return a string that represents the interval_t value of
   * the additional delay that needs to be applied to the outgoing message.
   *
   * <p>The returned additional delay in absence of after on network connection (i.e., if delay is
   * passed as a null) is NEVER. This has a special meaning in C library functions that send network
   * messages that carry timestamps (@see send_timed_message and send_port_absent_to_federate in
   * lib/core/federate.c). In this case, the sender will send its current tag as the timestamp of
   * the outgoing message without adding a microstep delay. If the user has assigned an after delay
   * to the network connection (that can be zero) either as a time value (e.g., 200 msec) or as a
   * literal (e.g., a parameter), that delay in nsec will be returned.
   *
   * @param delay The delay associated with a connection.
   */

As the NEVER value is a special value within reactor-c I am assuming this is a reactor-c issue and not an LF generation issue.

I did a inline check to replace NEVER with 0. Let me know if a different solution is preferred.

[SheaFixstars]

I missed that this code block was removed in the #323 pull. Will close this issue for now as it looks like a fix is already in the works.