Fix a bug in BAM random reader.

[alumi]

I found sometimes BAM random reading function like (cljam.io/read-alignment reader {:chr "chr1" :start 1 :end 10000000 :depth :deep}) returns fewer alignments than actual for certain BAM files.

The root cause was in cljam.bam.reader/read-alignments* function:

(let [...
      candidates (flatten (map (fn [[^Long begin ^Long finish]]
                                 (.seek ^BGZFInputStream (.reader rdr) begin)
                                 (read-to-finish rdr finish read-fn)) spans))]
  (if (= deep-or-shallow :first-only)
    (read-alignments-first-only rdr spans window read-fn)
    (filter window candidates)))

where cljam.bam.reader/read-to-finish returns a lazy sequence:

(defn- read-to-finish
   [^BAMReader rdr
    ^Long finish
    ^clojure.lang.IFn read-fn]
   (let [r ^BGZFInputStream (.reader rdr)]
    (when (and (not (zero? (.available r)))
               (> finish (.getFilePointer r)))
      (cons (read-fn rdr (.refs rdr))
            (lazy-seq (read-to-finish rdr finish read-fn))))))

Because (.seek ^BGZFInputStream (.reader rdr) begin) for all spans are called when flattened, read-to-finishreads alignments in the last span only. (+ the first one in all spans)

This problem is now fixed by seeking just before reading data
and its effect to performance of reading from single long span is limited.

[totakke]

LGTM