Fix dom_power issue on 32bit machine

[cyrilbouvier]

dom_power is used to compute n^l. In the code of master l is taken as an long.
This causes some issues on 32 bits machine (see for exemple linbox-team/linbox#293).

In this PR, I propose to switch the type from long to uint64_t.

First, I think an unsigned type should be used: in the code of dom_power, the argument l is immediatly cast to an unsigned long and the bit representation of this value is used to perform a double and add algorithm. It means that calling dom_power with a negative l is incorrect (except in the case where order(n) = ULONG_MAX+1).

Second, I think that using a type with a fixed size is clearer: just looking at the prototype of the function tells you the maximum possible value of the exponent.

This PR fixes linbox-team/linbox#293

[ClementPernet]

LGTM.