rewrite the endian conversion to fix a big-endian host

[kernigh]

(This might need a Jira ticket, but I don't have a Jira account. I will try to ask for a Jira account, but I'm not sure how.)

Client: py

Change htolell(dub) to htolell(transfer.t) in compact.h, because the conversion takes a long long, not a double. This matters on a big-endian host, where the conversion must swap bytes.

The big-endian letohll(n) failed because ntohl(n) is just n, doesn't swap n. Rewrite ntohll(n) and letohll(n) to use the same code with both big-endian and little-endian hosts. Expect the compiler to optimize away one of ntohll(n) or letohll(n) when it doesn't swap n.

• Did you create an Apache Jira ticket? (not required for trivial changes)
• If a ticket exists: Does your pull request title follow the pattern "THRIFT-NNNN: describe my issue"?
• Did you squash your changes to a single commit? (not required, but preferred)
• Did you do your best to avoid breaking changes? If one was needed, did you label the Jira ticket with "Breaking-Change"?
• If your change does not involve any code, include [skip ci] anywhere in the commit message to free up build resources.